Ranking is a central problem for many information retrieval applications. Apart from using traditional methods such as TF-TDF, VSM, LSI, BM25,... the application of machine learning techniques (ML) becomes a new trend in the field and has attracted great attention of research communities. Since 2007, LETOR - the benchmark dataset introduced by Microsoft - has been used widely in research papers and has become a useful tool for evaluating and comparing ranking methods and algorithms. In this paper, the authors present result on studying ML-based ranking methods tested on this dataset. the authors propose also an application of Genetic Programming to create ranking formulars with evaluation results on OHSUMED - a dataset in LETOR ver 3.0. The proposed method gives good results in comparison with existing methods.