Abstract: In traditional regress analysis,a good regression function is found in some function space based on known samples,such as polynomial function space,logarithm space and exponent space,When the sample number is certain the sample error will be big with the widening of function apace.In this paper,a method of sample error estimation is presented by probability measure and covering theory in a bouded function space.

Key words: sample error； covering theory； bounded function space； probability measure；