It depends on what the purpose of the tree is. Maximum likelihood and Bayesian methods require a lot of computation, and it may takes weeks to get the result. Neighbor-joining methods are far quicker and tend to give the same tree topology but perhaps with less accuracy of the branch lengths. For most data sets, the accuracy of the alignment, and the type of data uses (DNA vs amino acid, single gene vs multiple genes, nuclear genes vs mitochondrial, etc) is far more critical than the type phylogenetic analysis software used. Also, choosing the model of evolution and a few other factors are more important that the phylogeny method.