Most methods of personality scale construction have clear statistical disadvantages. A hybrid method (Darlington and Bishop, 1966) was found to increase scale validity more than any other method, with large item pools. A simple modification of the Darlington-Bishop method (algebraically and conceptually similar to ridge regression, but independently derived) is proposed and examined; it is intended to increase scale validity by reducing validity shrinkage. In a double cross-validation design, MMPI items were used to predict diagnostic category (schizophrenic vs. other) for 200 hospitalized patients. With one anomalous exception, the proposed technique produced scales more valid than those from the original Darlington-Bishop technique. (Author)