Previous researchers proposed a model to classify question items with artificial neural network approach that applies different feature method (Yusof & Chai, 2010). The model is trained using the scaled conjugate gradient learning algorithm. Several data processing techniques are applied to a feature set and then the content of a question is transformed into a numeric form called a feature vector. In order to perform text classification, three types of feature set are used i.e. whole feature set, the Document Frequency (DF) and Category Frequency-Document Frequency (CF-DF). A question item which consist of 274 questions were selected for processing. From the system, out of the three feature sets, DF reductions gave more efficient result with the combination of classification and convergence time