Future development of electric power systems must pursue anumber of different goals. The power system should be economically efficient, it should provide reliable energy supply and should not damage the environment. At the same time, operation and development of the system is influenced by a variety of uncertain and random factors. The planner attempts to find the best strategy from a large number of possible alternatives. Thus, the complexity of the problems related to power systems planning is mainly caused by presence of multiple objectives, uncertain information and large number of variables. This dissertation is devoted to consideration of the methods for development planning of a certain subsystem, i.e.the distribution network. The dissertation first tries to formulate the network planning problem in general form in terms of Bayesian Decision Theory. However, the difficulties associated with formulation of the utility functions make it almost impossible to apply the Bayesian approach directly. Moreover, when approaching the problem applying different methods it is important to consider the concave character of the utility function. This consideration directly leads to the multi-criteria formulation of the problem, since the decision is motivated not only by the expected value of revenues (or losses), but also by the associated risks. The conclusion is made that the difficulties caused by the tremendous complexity of the problem can be overcome either by introducing a number of simplifications, leading to the considerable loss in precision or applying methods based on modifications of Monte-Carlo or fuzzy arithmetic and Genetic Algorithms (GA), or Dynamic Programming (DP). In presence of uncertainty the planner aims at finding robust and flexible plans to reduce the risk of considerable losses. Several measures of risk are discussed. It is shown that measuring risk by regret may lead to risky solutions, therefore an alternative measure - Expected Maximum Value - is suggested. The general future model, called fuzzy-probabilistic tree of futures, integrates all classes of uncertain parameters (probabilistic, fuzzy and truly uncertain). The suggested network planning software incorporates three efficient applications of GA. The first algorithm searches simultaneously for the whole set of Pareto optimal solutions.The hybrid GA/DP approach benefits from the global optimization properties of GA and local search by DP resulting in original algorithm with improved convergence properties. Finally, the Stochastic GA can cope with noisy objective functions. Finally, two real distribution network planning projects dealing with primary distribution network in the large city and secondary network in the rural area are studied.