Types of cofactor independency for newly found oxidoreductases sequences are usually determined by experimental analysis. These experimental methods are both time-consuming and costly. With the explosion of oxidoreductases sequences entering into the databanks, it is highly desirable to explore the feasibility of selectively classifying newly found oxidoreductases into their respective cofactor independency classes by means of an automated method. In this study, we proposed a modified Chou's pseudo-amino acid composition method to extract features from sequences and the k-nearest neighbor was used as the classifier, and the results were very encouraging. When lambda = 48, w = 0.1, the areas under the ROC curve of k-nearest neighbor in 10-fold cross-validation was 0.9536; and the success rate was 92.0%, which was 3.5% higher than that of pseudo-amino acid composition. It was also better than all the other 7 feature extraction methods. Our results showed that predicting the cofactors of oxidoreductases was feasible and the modified pseudo-amino acid composition method may be a useful method for extracting features from protein sequences.