This report documents an investigation into two types of variables that might be useful in predicting flight grades in Navy primary flight training. The first set of predictor variables is largely psychomotor in origin and is part of the Computer-Based Performance Test Battery at the Naval Aerospace Medical Research Laboratory. The second set of variables is more cognitive in nature and arises from scores on the Aviation Selection Test Battery (ASTB) and a final grade in Aviation Pre-Flight Indoctrination (API), which is ground school prior to entering primary flight training. The motivation for this research is a joint effort with the Air Force designed to improve selection tests for military aviators. The emphasis in this report is on how to choose good linear regression models which use these variables to predict a criterion variable such as flight grade. In our present case, we have a total of 25 potential predictor variables. As a result, there is a rather large number of possible regression models. Our task is to pick some relatively small number of models that are "best" by some acceptable statistical criterion. The analysis revealed that models with a small number of predictor variables were much superior to models that included a large number of the 25 available variables. The best models consisted of two, three, and four predictor variables and possessed an R 2 of about .35. The single best model contained the final grade from API, a psychomotor tracking variable, and a score from one of the ASTB subtests. A prediction of the flight grade can then be made by averaging over the individual predictions of the single models. Using Bayesian model evaluation techniques, the averaging is carried out by weighting each individual model according to its posterior probability. Form Approved OMB No. 0704-0188 Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, aatherinq and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information including suggestions for reducing this burden, to