Abstract: In many clinical trials, the prediction of outcome following therapy requires analysis of two or more small groups of responders having a large number of simultaneously measured covariates, some of whose values may be absent. Prediction of individual outcomes in these groups is a severe statistical problem. This has motivated us to develop a suit-able approach for inference from such limited data. A new statistically-oriented prediction method, Optimized Independent Segment Voting (OISV), is presented for constructing a class-membership prediction function for such data sets. This "voting" prediction function is constructed based on the most informative and robust discrete segments of all covariate ranges, which are thus discretized.