Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Benzene formation in electronic cigarettes

  • James F. Pankow ,

    Affiliations Department of Chemistry, Portland State University, Portland, Oregon, United States of America, Department of Civil and Environmental Engineering, Portland State University, Portland, Oregon, United States of America

  • Kilsun Kim,

    Affiliation Department of Chemistry, Portland State University, Portland, Oregon, United States of America

  • Kevin J. McWhirter,

    Affiliation Department of Civil and Environmental Engineering, Portland State University, Portland, Oregon, United States of America

  • Wentai Luo,

    Affiliations Department of Chemistry, Portland State University, Portland, Oregon, United States of America, Department of Civil and Environmental Engineering, Portland State University, Portland, Oregon, United States of America

  • Jorge O. Escobedo,

    Affiliation Department of Chemistry, Portland State University, Portland, Oregon, United States of America

  • Robert M. Strongin,

    Affiliation Department of Chemistry, Portland State University, Portland, Oregon, United States of America

  • Anna K. Duell,

    Affiliation Department of Chemistry, Portland State University, Portland, Oregon, United States of America

  • David H. Peyton

    Affiliation Department of Chemistry, Portland State University, Portland, Oregon, United States of America



The heating of the fluids used in electronic cigarettes (“e-cigarettes”) used to create “vaping” aerosols is capable of causing a wide range of degradation reaction products. We investigated formation of benzene (an important human carcinogen) from e-cigarette fluids containing propylene glycol (PG), glycerol (GL), benzoic acid, the flavor chemical benzaldehyde, and nicotine.

Methods/Main results

Three e-cigarette devices were used: the JUULTM “pod” system (provides no user accessible settings other than flavor cartridge choice), and two refill tank systems that allowed a range of user accessible power settings. Benzene in the e-cigarette aerosols was determined by gas chromatography/mass spectrometry. Benzene formation was ND (not detected) in the JUUL system. In the two tank systems benzene was found to form from propylene glycol (PG) and glycerol (GL), and from the additives benzoic acid and benzaldehyde, especially at high power settings. With 50:50 PG+GL, for tank device 1 at 6W and 13W, the formed benzene concentrations were 1.9 and 750 μg/m3. For tank device 2, at 6W and 25W, the formed concentrations were ND and 1.8 μg/m3. With benzoic acid and benzaldehyde at ~10 mg/mL, for tank device 1, values at 13W were as high as 5000 μg/m3. For tank device 2 at 25W, all values were ≤~100 μg/m3. These values may be compared with what can be expected in a conventional (tobacco) cigarette, namely 200,000 μg/m3. Thus, the risks from benzene will be lower from e-cigarettes than from conventional cigarettes. However, ambient benzene air concentrations in the U.S. have typically been 1 μg/m3, so that benzene has been named the largest single known cancer-risk air toxic in the U.S. For non-smokers, chronically repeated exposure to benzene from e-cigarettes at levels such as 100 or higher μg/m3 will not be of negligible risk.


Electronic cigarettes (“e-cigarettes”) use an electrical resistance coil to vaporize mixtures of propylene glycol (PG), glycerol (GL), nicotine, and flavor chemicals. Vaporization of an e-liquid containing mostly PG and/or GL requires a temperature of ~190 to 290°C: when the ambient pressure is 1 atm, PG boils at ~190°C, GL boils at ~290°C, and PG+GL mixtures will boil between ~190 and ~290°C; the presence of other constituents besides PG and GL (such as water and flavor chemicals) will affect the boiling point. (The presence of significant percentages of other constituents (e.g., water and flavor chemicals) will affect the boiling point.) Temperatures higher than the boiling point of an e-liquid are possible in the coil zone if the rate of e-liquid delivery to the coil does not keep pace with the heat delivery rate: the vicinity of the coil becomes “dry”, and the heat delivery rate surpasses the rate at which heat is carried away by evaporated liquid as “latent heat”.

In general, e-cigarette aerosols tend to be simpler in composition than cigarette aerosols: “e-liquids” are a simpler starting matrix as compared to cigarette filler, and burning cigarettes have been reported to reach 900°C,[1]. Neverthless, multiple toxicants can form upon heating PG and GL.[26] Thermal dehydration of PG with loss of one water molecule gives acetaldehyde, and thermal dehydration of GL with loss of two water molecules gives acrolein [2, 6]. Significant amounts of formaldehyde are also possible.[3,4] Kim and Kim [7], using a PG+GL refill fluid (zero nicotine), an unnamed refillable tank device operated, and unspecified settings, reported finding benzene (a known human carcinogen [8,9]) in e-cigarette aerosols at 87.5 μg/m3. McAuley et al.[10], however, using a simple draw-activated device, reported that benzene was mostly “not found”.

Dehydration of GL to benzene has been observed [11], and in e-cigarettes a simple dehydration stoichiometry could be PG + GL = benzene + 5 H2O (Fig 1A). A second route to benzene in e-cigarettes is decarboxylation of benzoic acid (Fig 1B), and benzene has been known to form when benzoic acid is used as a preservative in beverages.[12] (Benzoic acid has been found by our laboratory in 14 out of 150 e-liquid refill products at levels estimated to be in the range 0.02 to 2 mg/mL, and benzoic acid is an acknowledged ingredient in e-liquids in the JUUL product line.[13]) For a third route to benzene, many aromatic aldehydes are major e-liquid flavor additives, including benzaldehyde (for “cherry”), vanillin, and ethyl vanillin: aldehyde levels as high as several percent (by mass) have been found.[14] Every aldehyde can be oxidized to its corresponding carboxylic acid, which may then undergo decarboxlation. Thus, oxidation of benzaldehyde can give benzoic acid, and therefore, benzene (Fig 1C). For a fourth route to benzene, in what amounts to abiotic fermentation, an aldehyde can undergo redox disproportionation to form a mix of the corresponding alcohol and the acid, and the latter may then undergo decarboxylation. (The acid is more oxidized then the aldehyde, and the alcohol is less oxidized than the aldehyde.) With benzaldehyde, a mix of benzoic acid and benzyl alcohol can then be formed (Fig 1D). (The disproportionation of an aldehyde lacking an “alpha-position” hydrogen atom is the Cannizzaro reaction, which is base-catalyzed (possibly then, by nicotine).)

Fig 1.

Formation of benzene by four mechanisms: a. dehydration according to GL + PG– 5 H2O, with cyclization (note: individually, propylene glycol alone and glycerol follow different stoichiometries); b. decarboxylation of benzoic acid; c. oxidation of benzaldehyde to benzoic acid, followed by decarboxylation (dashed arrow—-> indicates that the exact reaction stoichiometry is not provided); and d. disproportionation (Cannizzaro reaction) of benzaldehyde to form benzoic acid + benzyl alcohol.

Herein we describe measurements of gas-phase benzene in e-cigarette aerosols from three types of e-cigarette: a non-refillable e-cigarette (JUULTM), and two variable-power, tank-type devices. For experiments with the tank devices, the fluids used were prepared in the laboratory from PG, GL, benzoic acid, benzaldehyde, and/or nicotine (see Table 1 for compositions). The power settings used for the tank units ranged from “recommended” to beyond. The higher settings were used because they: 1) were accessible by normal use of the devices; 2) may not be “distasteful” to absolutely every user in every use circumstance; 3) will certainly be encountered by users experimenting with settings (as innumerable postings on social media attest); and 4) provide useful information regarding the potential for toxicant formation in e-cigarettes.

Table 1. Benzene and Total Particulate Matter (TPM) in E-Cigarette Aerosols Generated Using Two Devices and Different Lab-Prepared Fluids Based on Propylene Glycol (PG) and/or Glycerol (GL).

When Together, PG and GL Combined in Equi-volume Amounts. 13C-Labelled Compounds Only Present As Indicated.

Materials and methods

Chemicals and e-cigarette devices

Fully 13C-labelled PG and fully 13C-labelled GL were obtained from Cambridge Isotopes Laboratory (Tewksbury, MA). Non-labeled PG and GL and standard chemicals were obtained from Sigma-Aldrich Inc. (St. Louis, MO). Determination of benzene levels in the e-cigarette aerosols was by cartridge-based adsorption/thermal desorption (ATD) followed by gas chromatography with detection by mass spectrometry (GC/MS). Three different e-cigarette devices were used in the ATD determinations: 1) JUULTM personal vaporizer and refill cartridges (“pods”) (Pax Inc., San Francisco, CA) in four different flavors (tobacco, mint, fruit, and crème brûlée) were purchased online in May 2016 (the fluids were analyzed in this study to determine the levels of benzoic acid and nicotine); 2) EVODTM tank-type atomizer (Kangertech, Shenzhen, China) with 1.8 ohm resistance single horizontal coil and silica wicking material, purchased online in July, 2016; and 3) Subtank NanoTM V.1 (Kangertech) tank-type atomizer with 1.2 ohm resistance single vertical “OCC” single coil and cotton wicking material, purchased online in July, 2016. The JUULTM system has no user options other than flavor of the cartridge selected. For the EVODTM and Subtank NanoTM devices: 1) the “recommended” settings were 6W and 10 to 26W, respectively; 2) each replicate aerosol sample for gaseous benzene determination proceeded using a clean tank and a new coil so that any run-to-run changes in benzene production would not be caused by “aging” of the coil etc.; 3) at every wattage setting tested, sample collection was begun 2 h after “conditioning” the new coil; 4) conditioning occurred by taking six 50 mL puffs at 6 W for the EVOD device and 13W for Subtank NanoTM device respectively.


For all three devices, each ATD sample was taken as three or six 50 mL puffs without power (blanks) or with power (aerosol samples) with a puff duration of 5 s and puff-to-puff interval of 60 s. (Regarding the selected puff duration, Hua et al. [15] used a data mining exercise for 64 different ENDS users on YouTube to obtain an average puff duration of 4.3 seconds ± 1.4 seconds (SD) for men, and an average of 4.0 seconds ± 0.8 seconds (SD) for women.)

Each ATD sample comprised an average benzene level for six puffs. For the JUULTM device, blank samples were collected by drawing lab air through an electrically disconnected cartridge. For e-cigarette aerosol sampling, the JUULTM battery was then connected and device activation occurred automatically as a disposable 60 mL syringe was used to manually pull a puff through a 0.45 μm pore size 28 mm diameter (Phenomenex Inc., Torrance, CA) glass fiber/cellulose acetate (GF/CA) filter (to remove the aerosol droplets), followed by a single ATD gas sampling cartridge containing 100 mg of 35/60 mesh Tenax TA and 200 mg of 60/80 mesh Carbograph 1 TD (Camsco Inc., Houston, TX). All four different JUULTM flavors were tested. The direct “butt” connections to the filter were held in place using short pieces of flexible 0.125 in. i.d. polyvinyl chloride (PVC) tubing. The benzene levels obtained are minimum values because of the possibility of some small sorptive loss to the PVC pieces; the PVC pieces were replaced after each sample. For the method used (indoor sampling), blank levels corresponded to ~5 μg/m3. Final values given are blank-corrected; values not significantly above the blank level are reported as “not detected” (ND).

For the two tank units, the device used for drawing e-cigarette puffs involved a programmable syringe pump. As above, direct “butt” connections were made using short pieces of PVC tubing. Blanks were obtained without activating the power. For vaping, flow for each sample was sequentially drawn through: 1) a GF/CA filter as above; 2) single ATD gas sampling cartridge as above; a 3) three way “T” valve connected to a 1 L Tedlar bag (Model 24633 Supelco Inc. (Bellefonte, PA); and 4) a syringe pump (Model NE-1010, New Era Pump Systems Inc., Farmingdale, NY). Minimal “butt” connections were made as above. For verification of the total sample volume, the puffs were exhausted from the syringe pump back through the three way valve and into the Tedlar bag. Again, final values are blank-corrected, and values not significantly above the blank level are reported as ND.

For confirmation of the formation of benzene in the e-cigarette aerosols by nuclear magnetic resonance (NMR) spectroscopy, aerosol samples were created using an EVODTM device (as above) with 50:50 PG+GL containing benzoic acid at 1% by weight. Aerosol was generated at 14 W (2 ohms resistance) using five 50 mL puffs, each drawn over five seconds with a puff interval of 1 min. In an approach similar to that described by Jensen et al. [4], the aerosol was drawn into a 1 mL septum vial containing 600 μL of DMSO-d6. The flow inlet to the vial was an 18 gauge needle, as was the flow outlet. (Gas-phase benzene will be captured efficiently by this method.) Samples were run using a 600 MHz NMR spectrometer.

Mass of e-liquid vaped

Values for mg of e-liquid vaped per puff (and for the resulting TPM concentration) were estimated based on weight loss of the e-cigarette unit for each run.

Gas-phase benzene

For all the e-cigarette aerosol samples, >89% of the benzene can be deduced to have been in the gas phase as follows: 1) As discussed by Pankow et al.[16,17] the percent of a compound in the gas phase of an aerosol is given by P(%) = 100%/(1+KpTPM) where Kp (m3/μg) is the gas-to-particle (i.e. droplet) partition coefficient, and TPM is the suspended “total particulate matter” level for the aerosol. 2) It can be estimated that Kp ≈ 10−9.7 m3/μg for benzene at 20°C for partitioning to e-cigarette aerosol particles (as based on the approach of Pankow et al.,[16,17] using a benzene vapor pressure of 0.099 atm at t = 20°C,[18] an activity coefficient for benzene (ζbenzene) in glycol solutions of ~14 [19], and an average molecular weight of ≈ 84 g/mol for a 50:50 by volume PG:GL mixture). 3) For all the aerosols created here, TPM ≤ 108.8 μg/m3 (= aerosolized liquid mass/total puff volume). Then P(%) ≥ 100%/(1+10−9.7108.8) = 89%.

ATD cartridge analyses by GC/MS

For each standard sample for method calibration, an ATD cartridge was charged with 2 or 4 μL of a 1 to 100 ng/μL solution of benzene in methanol, followed by a 50 mL/min flow of nitrogen for 10 min. Each ATD sample cartridge and each standard ATD cartridge was thermally desorbed using a TurboMatrix 650 ATD unit (PerkinElmer, Waltham, MA). Prior to desorption, the ATD unit automatically amended each cartridge with 20 ng of fluorobenzene as the internal standard compound. Each ATD cartridge was thermally desorbed for 10 min at 285°C with helium desorption flow of 40 mL/min and split flow of 20 mL/min. The desorption stream was trapped at -10°C on the intermediate “air monitoring trap” (ATP). The ATP was then thermally desorbed at 295°C and 25 psi constant pressure helium with a split flow of 8 mL/min for 4 min. The non-split portion of the desorption gas stream passed onto the GC column which was mounted in an Agilent (Santa Clara, CA) 7890A GC. The GC was interfaced to an Agilent 5975C MS operated in electron impact ionization mode. The MS scan range was 34 to 400 amu. The electron multiplier voltage was 1400 V. The fused silica capillary GC column was a model Rxi-624Sil MS (Restek Inc., Bellefonte, PA) of 30 m length, 0.25 mm i.d., and 1.4 μm film thickness.

JUULTM pod analyses by LC/UV

JUULTM pods were opened to obtain the e-liquid for determination of the benzoic acid and nicotine concentrations. Aliquots of 100 μL were withdrawn, diluted 1:100 with methanol, and filtered with a syringe-mounted PVDF (polyvinylidene fluoride) filter, 13 mm diameter, 0.22 μm pore size, obtained from Thermo Fisher Scientific, Inc. (Waltham, MA). Analyses proceeded by high performance liquid chromatography (HPLC) using an injection volume of 20 μL, a Waters Corp. (Milford, MA) Model 1525 binary solvent delivery module (with Rheodyne 7725i injector), a Discovery™ C-18 column (250 × 4.6 mm × 5 μm, Supelco Inc.) at 40°C, and a Waters Corp. Model 2996 photodiode array detector. The mobile phase composition was maintained isocratic, 40:60 methanol:water amended with 0.1% trifluoroacetic acid (1 mL/min). The detector wavelength for benzoic acid was 228 nm, and 259 nm for nicotine. The calibration ranges for the injected standards were 0 to 1 mg/mL for benzoic acid, and 0 to 0.1 mg/mL for nicotine.

Results and discussion

Benzoic acid and nicotine levels in JUULTM pod fluids

The concentrations of benzoic acid and nicotine in the JUULTM pod fluids were found to be 44.8 ± 0.6 and 61.6 ± 1.5 mg/mL respectively, corresponding to a benzoic acid/nicotine molar concentration ratio of 0.97 to 1). For comparison, as noted above, analyses in our laboratory have indicated the presence of benzoic acid in 14 commercial refill e-liquids at levels estimated to be in the range 0.02 to 2 mg/mL.

Benzene confirmation by NMR

Benzene formation (12C6) was confirmed by NMR for e-cigarette aerosol collected in DMSO-d6 and generated with the EVODTM device operated at 14 W (2 ohms resistance) using 50:50 PG+GL (both 12C3) containing benzoic acid (12C6) at 1% by weight. The expected strong singlet peak in DMSO-d6 at 7.37 ppm [20] was observed; addition of authentic benzene caused the peak to increase without introduction of other peaks.

Formation of 13C6 benzene From 50:50 PG+GL (both 13C3)

Formation of benzene with all carbons 13C (i.e., 13C6 benzene, MW = 84) was confirmed to occur from 50:50 PG+GL (both 13C3) (see Table 1) at levels consistent with the analogous runs with 50:50 PG+GL (both 12C3) (see Table 1).

Gas-phase benzene levels measured in e-cigarette aerosols

Gas-phase benzene was not found above blank levels in any of the JUULTM samples for any of the four flavors. For the other two devices, benzene gas-phase concentrations and deliveries (as μg benzene per g e-liquid vaped) are summarized in Fig 2, with other details provided in Table 1. Benzene formation was found to be very strongly dependent on device. With 50:50 PG+GL, for the EVOD device at 6W and 13W, the mean concentration values were 1.7 and 750 μg/m3, respectively. For the Subtank Nano device, at 6W and 25W, the mean values were remarkably lower, namely ND and only 1.5 μg/m3, respectively.

Fig 2. Benzene levels in e-cigarette aerosols generated with two different devices, different power levels, and 50:50 propylene glycol:glycerol with and without nicotine, benzoic acid, and/or benzaldehyde.

Significant formation from benzoic acid and benzaldehyde was observed for both the EVOD and Subtank Nano devices at the high power settings used. With benzoic acid and benzaldehyde at 9 to 10 mg/mL, for the EVOD device, values at higher power levels were as high as 5000 μg/m3. Remarkably, the values at the higher power levels for the Subtank Nano device were much lower, ≤ ~100 μg/m3.

Comparisons with benzene levels in ambient air and cigarette tobacco smoke

For tobacco smoke from “regular” cigarettes, the 1999 Massachusetts Benchmark Study [21] reported an average benzene delivery of 86 μg/cigarette. Assuming a total puff volume of ~400 mL/cigarette for the smoking protocol used [22], such deliveries correspond to a smoke benzene concentration of ~200,000 μg/m3, and so a much higher risk from benzene for chronic use of tobacco cigarettes as compared to e-cigarettes. However, median ambient air concentrations of benzene in locations in the U.S. in 2013 were ~1 μg/m3 [23]. Levels such as these resulted in benzene being named by the 2002 National-Scale Air Toxics Assessment (NATA) as the largest single known cancer-risk air toxic in the U.S.[24]. It can therefore be concluded for non-smokers that chronically repeated exposure to benzene from e-cigarettes at levels such as 100 μg/m3 will not be of negligible risk.

Supporting information

S1 Text. Information on e-cigarettes used.



NIH and FDA supported this work via award R01ES025257. In particular, the work reported in this publication was supported by NIEHS and the FDA Center for Tobacco Products (CTP). The content is solely the responsibility of the author and does not necessarily represent the official views of the NIH or the Food and Drug Administration.

Author Contributions

  1. Conceptualization: JFP DHP WL KJM KK JOE RMS AKD.
  2. Investigation: KK KJM WL JOE AKD.
  3. Project administration: JFP DHP.
  4. Supervision: JFP DHP RMS.
  5. Validation: KK KJM WL JOE AKD DHP JFP.
  6. Writing – original draft: JFP WL KJM KK JOE DHP.
  7. Writing – review & editing: JFP DHP.


  1. 1. Baker RR (1974) Temperature distribution inside a burning cigarette. Nature 247:405–406.
  2. 2. Nef JU (1904) Dissociationsvorgänge in der Glycol-Glycerinreihe. Justus Liebigs Ann Chem 335:247–279.
  3. 3. Kosmider L, Sobczak A, Fik M, Knysak J, Zaciera M, Kurek J, et al. (2014) Carbonyl compounds in electronic cigarette vapors: effects of nicotine solvent and battery output voltage. Nicotine Tobacco Research 16:1319–26. pmid:24832759
  4. 4. Jensen RP, Luo W, Pankow JF, Strongin RM, Peyton DH (2015) Hidden formaldehyde in e-cigarette aerosols. New England J. Medicine 372:392–394.
  5. 5. Sleiman M, Logue JM, Montesinos VN, Russell ML†, Litter MI, Gundel LA, et al. (2016) Emissions from electronic cigarettes: key parameters affecting the release of harmful chemicals. Environ Sci Technol 50: 9644–9651. pmid:27461870
  6. 6. Jensen RP, Strongin RM, Peyton DH. Solvent chemistry in the electronic cigarette reaction vessel. Scientific Reports 7:42549. pmid:28195231
  7. 7. Kim Y-H, Kim K-H (2015) A novel method to quantify the emission and conversion of VOCs in the smoking of electronic cigarettes. Scientific Reports 5:16383, 1–9; pmid:26553711
  8. 8. IARC (International Agency for Research on Cancer). Monographs on the Evaluation of Carcinogenic Risks to Humans Volume 100F (2012). Accessed August 16, 2016.
  9. 9. U.S. EPA, Integrated Risk Information System (IRIS) Chemical Assessment Summary U.S. Environmental Protection Agency National Center for Environmental Assessment, Benzene; CASRN 71-43-2, Section II. Carcinogenicity Assessment for Lifetime Exposure (Last Revised—01/19/2000). Accessed August 16, 2016.
  10. 10. McAuley TR, Hopke PK, Zhao J, Babaian S (2012) Comparison of the effects of e-cigarette vapor and cigarette smoke on indoor air quality. Inhalation Toxicology 24:850–857. pmid:23033998
  11. 11. Hoang TQ, Zhu X, Danuthai T, Lobban LL, Resasco DE, Mallinson RG (2010) Conversion of glycerol to alkyl-aromatics over zeolites. Energy and Fuels 24:3804–3809.
  12. 12. U.S. Food and Drug Administration. Data on Benzene in Soft Drinks and Other Beverages. Accessed August 15, 2016.
  13. 13. JUUL, Accessed August 15, 2016.
  14. 14. Tierney PA, Karpinski CD, Brown JE, Luo W, Pankow JF (2015) Flavour chemicals in electronic cigarette fluids. Tob Control.
  15. 15. Hua M, Yip H, Talbot P (2013) Mining data on usage of electronic nicotine delivery systems (ENDS) from YouTube videos. Tobacco Control 22: 103–106. pmid:22116832
  16. 16. Pankow JF, Luo W, Tavakoli AD, Chen C, Isabelle LM (2004) Delivery levels and behavior of 1,3-butadiene, acrylonitrile, benzene, and other toxic volatile organic compounds in mainstream tobacco smoke from two brands of commercial cigarettes. Chemical Research in Toxicology 17:805–813. pmid:15206901
  17. 17. Pankow JF (2017) Calculating compound dependent gas-droplet distributions in aerosols of propylene glycol and glycerol from electronic cigarettes. Journal of Aerosol Science 107: 9–13.
  18. 18. Dortmund Data Bank, DDBST GmbH.
  19. 19. Opris I. (1981) Determination and interpretation of activity-coefficients at infinite dilution of some hydrocarbons in terminal dihydroxy alcohols. Revista de Chimie (Bucharest) 32 234–238, 1981.
  20. 20. Gottlieb HE, Kotlyar V, Nudelman A (1997) NMR chemical shifts of common laboratory solvents as trace impurities. J Org Chem 62: 7512–7515. pmid:11671879
  21. 21. Borgerding MF, Bodnar JA, Wingate DE. The 1999 Massachusetts Benchmark Study. Final Report (2000). A Research Study Conducted after Consultation with the Massachusetts Department of Public Health, Department of Health, Massachusetts.
  22. 22. Pankow JF, Watanabe K, Toccalino P, Luo W, Austin D (2007) Calculated cancer risks for conventional and ‘‘potentially reduced exposure product” cigarettes. Cancer Epidemiology, Biomarkers, and Prevention 16:584–592.
  23. 23. U.S. Environmental Protection Agency. EPA's Report on the Environment; Ambient Concentrations of Selected Air Toxics. Exhibit 3. Ambient benzene concentrations in the U.S., 2003–2013. Accessed September 2, 2016.
  24. 24. George BJ, Schultz BD, Palma T, Vette AF, Whitaker DA, Williams RW (2011) An evaluation of EPA’s National-Scale Air Toxics Assessment (NATA): Comparison with benzene measurements in Detroit, Michigan. Atmospheric Environment 45:3301–3308.