Here are my estimates for Data Set V from BCI Competition III (seond algorithm). I an elman neural network to train on datasets 1and 3. The second dataset was used as validation to stop training due to overfitting. The elman network had three layers, first layer used 96 neurons with tansig TF, second layer had 60 neurons with purelin tf, and last layer had 3 neurons with logsig tf. The network also had 15 delayed input weights. Average percent correct was 70% on validation set.