Referee 1's Review:

**********
N.B. These are the comments made by the referee when reviewing an earlier version of this paper. Prior to publication the manuscript has been revised in light of these comments and to address other editorial requirements.
**********

Resting-state networks (RSN) driven by spontaneous, low-frequency fMRI signal changes have recently been studied by different research groups. Both informed data analysis approaches such as seed-based ROI functional connectivity analysis as well as model-free data driven techniques such as independent component analysis have previously been applied to resting-state fMRI data sets. To date, up to 10 independent resting-state networks have been described in the literature (Damoiseaux et al., PNAS 2006). In this interesting study, van den Heuvel and co-workers adds to the current literature by applying the normalized cut clustering algorithm to investigate the spatial characteristics of RSNs in the brain. The novel normalized cut clustering algoritm yielded seven RSN in the resting human brain.

1) First, the authors should be commended for their large sample size used in their analysis. Their results are in part in in agreement with previous model-free investigations of resting-state networks. The default mode (fig. 3a), left (fig. 3b) and right (fig. 3c) dorso-lateral fronto-parietal networks as well as the insular/ mid-cingulate cortex netework are very similar to the networks previously described by Damoiseaux. However, whereas both DeLuca (De Luca et al., Neuroimage, 2006) and Damoiseaux have shown data that support the idea that the sensorimotor cortex and the visual cortex belong to separate resting-state networks, the present results assign them to a common RSN. This finding is also at odds with other studies that have focused on the visual and motor system during rest. Moreover, when looking at the networks presented in Supplementary figure S1 one gets the impression that cluster "f" strongly resembles the default mode network with activity in the precuneus/ posterior cingulate cortex and the medial prefrontal cortex. However, this is not appearent in the presentation in Figure 3. Moreover, wheras the Damoiseaux study could identify separate RSN for primary versus extra-striatal visual areas (c.f. Fig 2A & E in Damoiseaux et al., PNAS 2006), the present study collapses them into a common RSN together with motor/sensory areas (Fig 3d). Taken together, I believe that the above mentioned discepancies between the present study and the previous literature warrants a re-analysis of the authors data. In this re-analysis, I would suggest that the authors try different thresholds (see also comment below) for the connectivity graph cut-off threshold, i.e. the voxel pair-wise correlation coefficient threshold) and investigate what effects different settings will cause on the final clustering results, similar to what the authors have done for other user settings displayed in supplementary figures S1 and S2.

2) The authors must describe the Ncut clustering algorithm in more a lot more detail. Exactly how was the clustering shown in Figure 1 A2 done? Moreover, exactly how was the optimization of the cluster parameters in step B2 done? The reader should not have to consult an IEEE paper to understand the clustering procedure done by the authors.

3) The authors state in the abstract and the discussion that the used normalized cur group clustering algorithm and the calculation of the number of resting state networks estimated from the EPI data is "fully automatic". I think this is an overstatement, since the authors analysis required intervention from the user at several different stages. As pointed out by the authors, the user must supply connectivity graph cut-off threshold (step A1, set to 0.4 by the authors), number of expected clusters (step A2) as well as a graph complexity cut-off threshold (step B2). Although the algorithm's sensitivity to the latter two was was addressed by the authors (figure S1 and S2), they nevertheless remains to be specified by the user. Hence, I suggest that these short-comings should be more explicitly described by the authors in the manuscript.

4) Results, page 11: The authors describe the similarities with other studies but fail to properly describe the differences between their results and the previous literature. Please elaborate on this issue.

5) How was the normalized cut clustering algoritm implemented? Which software packages was used (Matlab, IDL, C++?). Is the implementation of the algorithm freely available to other investigators?

6) Discussion, page 12, last sentence: I do not agree with the claim that the graph representation of the network will provide the reader with information regaring both structural and functional properties of RSN. First, the graph representation is only used internally by the normalized cut algorithm and is not directly accessible to the ser. Second, theclustering algorithm will provide the user with nformation regarding the spatial characterstics of distributed RSN, but the functional relevance of these networks must be infered from other sources, such as conventional task-related neuroimaging experiments, lesions studies etc. Please rephrase this sentence.

7) Discussion, page 13, last sentence. The authors write "Our results support the idea of a default state of the brain and the formation of functional dependent resting-state networks in the brain at rest". I disagree. What the authors have shown is that they found experimental support for 7 spatially distinct RSN in the human brain, including the so called default mode network. This is not the same thing as saying that there is support for "the idea of a default mode in the brain". Moreover, the question whether they are functional dependent or independent remains to be investigated. Please rephrase.

8) Methods section, page 5: Was there a particular reason to realign all EPI volumes to the last scan and not the first in each functional image series?

9) References 27, 35, 36, and 40 are incomplete (issue and page number information is missing). Please add this information.