It is nice to see a SVM being evaluated for prediction of type III secretion signals but the results are preliminary, as the authors state themselves. It's annoying that contrary to what is said in the article (page 2), the training sets are not available. For this, the paper cannnot be judged in detail. It is hard to agree on the conclusion that there is a "low number of false positives". For instance, the fraction of predicted type III-secreted proteins in Xanthomonas (having a T3SS) is 13.7 % while that for Xylella (a closely related species without T3SS) is 13.8 %. This does not support the conclusion that this algorithm might be helpful to select candidates for an experimental survey.