Haplotypes, rather than genotypes are required in some disease susceptibilities and drug response tests.However, it is both time-consuming and expensive to obtain haplotypes experimentally. Therefore usually genotype data are collected in the laboratory at first, then, haplotype data are inferred from them resorting to some computational approaches. Different from Clark's well-known haplotype inference method, Gusfield and Wang et al.proposed a new model according to the maximum parsimony principle. It tries to find a minimum set of haplotypes that can explain the genotype samples. This parsimony model overcomes some weaknesses of Clark's method. For the parsimony this paper presents model a polynomial time greedy algorithm and a compound algorithm that combines the greedy policy with the branch-and-bound strategy in a uniform framework. Compared with the original complete algorithm proposed by Wang et al., the greedy approximation algorithm runs much faster, and in the meanwhile, produces relatively higher accurate results. The compound algorithm is also a complete algorithm.Simulation results show that it is much more efficient and can be applied to instances of much larger scales than the original complete algorithm.