The paper introduces a new concept of computation tree of a logic program that will be used in reasoning about programs. Three types of transformations improving the structure of logic programs are described. There are two natural measures of complexity suggested by computation trees, namely, the number of nodes called by recursion and the maximal number of and/or alternations on a branch. Both measures are shown to collapse, or more precisely, it is shown how every logic program can be transformd to a program computing the same function having a computation tree with at most one called node and at most two alternations on every branch. This possibility allows to restrict the class of logic programs which must be considered. There are compared theoretical results with the attempts to develop a practical methodology for constructing logic programs.