Three different approaches to shape recognition were evaluated. The recognition powers were demonstrated with real irregular objects and computational aspects were discussed. The chain code histogram is a simple descriptor which is fast to calculate. However, it has some limitations which affect its generality. The pairwise geometric histogram is a very powerful descriptor which can be used especially with polygonal objects that can even be overlapping. It is, however, very computing intensive and requires much memory. The set of five simple shape descriptors provides a good compromise between recognition power and computational complexity. It performs equally well with the PGH but requires less computation time and memory.