—Low-density parity-check (LDPC) codes are gaining interest for high data rate applications in both terrestrial and spatial communications. They can be designed and studied through a bipartite graph whose characteristics affect the performance. This paper proposes a low-complexity method to improve the performance of LDPC codes by selectively removing some cycles from the associated bipartite graph. The method is based on a modified version of the breadth first search (BFS) algorithm that we call modified BFS (MBFS), which is applied to find cycles, and a greedy procedure to eliminate them. Throughout the paper we will give a detailed description of the algorithm proposed and analytically study its complexity. Simulation results show that this girth conditioning method applied to some classes of codes, whose structure allows further optimization, can lead to a significative complexity reduction and a performance improvements with respect to other methods.