Gene expression data often need to be classified into classes or grouped into clusters for further analysis, using different machine learning techniques and an important pre-processing step is feature selection (FS). The aim of this study is to investigate the stability of some diverse FS methods on a plethora of microarray gene expression data. This experimental work is broken into three parts. Step 1 involves running some FS methods on one gene expression dataset to have a preliminary assessment on the similarity, or dissimilarity, of the resulting feature subsets across methods. Step 2 involves running two of these methods on a large number of different datasets to investigate whether the results produced by the methods are dependent on the features of the dataset: binary, multiclass, small or large dataset. The final step explores how the similarity of selected feature subsets between pairs of methods evolves as the size of the subsets are increased. Results show that the studied methods display a high amount of variability in terms of the resulting selected features. The feature subsets differed both inter- and intra- methods for different datasets. The reason behind this is not clear yet and is being further investigated. The final objective of the research, that is to define how to select a FS method, is an ongoing work whose initial findings are reported herein.