Based on the features of the genetic algorithm and radial distribution structure， a novel hybrid algorithm for reactive power planning optimization in distribution networks is proposed in which the determinate and stochastic features are organically combined. By adopting a method of dynamic upper limit to node compensation out of consideration for both economic benefits and voltage quality， the problem of determining the upper limit and the total number of candidate nodes beforehand by experience is solved and the resulting initial population could become mostly feasible. For the voltage quality problem， a strategy for optimization by stages is proposed in which the adaptive adjustment mechanism of penalty factors is used. Moreover， simulated annealing， an optimization crossing method and a dynamic disaster mechanism are employed in the proposed hybrid algorithm in addition to a dynamic threshold verification mechanism of node maximum compensation capacities. The validity and effectiveness of the proposed algorithm is demonstrated by an IEEE 33-bus test system and several real radial distribution networks.This work is supported by Technology Science Innovation Fund of Ministry of Science（No. 06C26225110507）.