Estimating dependences on the basis of empirical data has been, and will probably remain, a central problem in applied analysis. This problem is a mathematical interpretation of one of the basic questions of science: how to extract the existing law-like relationship from scattered data. The simplest attack on this problem is to construct (estimate) a function from its values at certain points. Here we will formulate some general principles of estimating a functional dependence, and then develop an algorithm for the estimation using these principles.