"problem evaluating classifier train and test set are not compatible" for KDD99 test'

Hi All,

I have a question for KDD99 %10 training and testing dataset.

1. I discretizated and do the feature selection for the test data then save it as .arff file.
2. I discretizated, do the feature selection and classified the train data.
3. In the classification tab I chose my test set in the supplied test section.
4. I did the Re-evalution (In the right click options)
5. I got this, ''problem evaluating classifier train and test set are not compatible''

What could be the possible reason? I googled and read lots of things but, it`s weird.
Training data contains: 494021 instances
Test data contains: 331029 instances

@attribute service {vmnet.....} and @attribute label {back,saint..........} includes different values but test data contains the training data`s values (in the attribute values).

Is there any obligation that all values should be the same in the attribute values?

Any feedback appreciated..

Thanks in advance.