We studied the possibility of the application of known risk minimization conception to the solution of nonlinear inverse gravity problems when the problem of choice is defined under conditions of indeterminacy. In accordance to this conception, the best feasible solution is the one which provides the minimum of the error expectation for interpretation results. We suggest the effective algorithm to construct the representative sample of feasible variants of the interpretation, one of them will be selected as the best. The important advantage of the risk minimization method is the possibility of the approximate estimator of the similarity between the mathematical model and the true object. Results of the multivariant computational experiment approving the efficiency of these estimators are presented in the article.