The first tests of algorithm SPA [1] were carried out on a task of medical diagnostics. Three images were submitted by training set of 250 realizations in 17-th dimansion  space. It was required to find most informative subspace of dimension 3 and 6. In the beginning by method of full running over found the optimum decisions of this task.        Then the same best decisions were found by a method SPA. Has appeared, that the time of search of the decision by a method SPA is less than time of full running over: for n = 3 in 5 times and for n = 6 - in 40 times. This prize on time quickly grows with increase g and n.         In more complex cases, where full running over was not possible, quality of a subsystem of attributes chosen by a method SPA, were compared to quality of initial system from g of attributes. As a rule, for acceptable time the algorithm SPA chose a subsystem, which on informability poorly conceded to initial system, that allows to consider the chosen subsystem close to optimum.         On the same examples the algorithm SPA showed the best results, than algorithms Del [2] and Ad [3].  The literature: 1. Lbov G.S. Methods of processing  the different scale  experimental   data. Published by  "Science" the Siberian branch, Novosibirsk, 1981.  2. Merill T., Green O.M. On the effectiveness of receptors in recognition   systems.- JRE Trans. Inform. Theory, 1963, vol. JT-9, p.11-17.    3. Barabash Ju. L., Varsky B.V., etc. Automatic pattarn recognition. Published by KVAIU, Kiev, 1964.