One of the aims of the Configural Frequency Analysis (CFA) is the identification of symptom configurations as types or anti-types. Following the pattern of an efficient regression algorithm of Cierzynski and von Weber the authors developed a gradient method, which minimizes iteratively the total c² of a contingency table, reducing by a small amount the frequency of the most suspicious cell in the step of iteration (in the case of an anti-type the frequency will be increased). The final result is a table, which fulfills perfectly the hypothesis of independence. The expectation values one can calculate from this table are known as Victor-expectation values in the literature. With these values and the original cell frequencies the trustworthy small-group test of Dunkl and von Eye is performed and then confirmed by Holm’s procedure. With a Bayesian ansatz (types are hidden by preference in highly frequented cells) a further improvement of the results has been reached. Numerical simulations are showing the correctness of the method. A comparison with the results of the "new approach" of Kieser and Victor is showing a large measure of conformity.