An implementation of a novel strategy for cluster geometry optimization, using a stochastic method, is applied. This algorithm is based on the spirit of Big Bang theory. The strategy consists on a three-step procedure; a semiempirical optimization of the 1000n super compressed starting geometries, followed by two DFT optimizations of the resulting structures. The methodology is applied to the study of alkali metal clusters (Li n with 3 n 20, Na n with 3 n 13, and K n with 3 n 10) structures and properties. With this new technique, we were able to find new isomers. The structures of alkali clusters show a large number of pentagonal pyramids and tetrahedral subunits. Most important electronic properties have been discussed and compared with different calculations and experimental data.