Molecular diagnostics tools provide specific data that have high dimensionality due to many factors analyzed in one experiment and few records due to high costs of the experiments. This study addresses the problem of dimensionality in melanoma patient antibody display data by applying data mining feature selection techniques. The article describes feature selection ranking and subset selection approaches and analyzes the performance of various methods evaluating selected feature subsets using classification algorithms C4.5, Random Forest, SVM and Naïve Bayes, which have to differentiate between cancer patient data and healthy donor data. The feature selection methods include correlation-based, consistency based and wrapper subset selection algorithms as well as statistical, information evaluation, prediction potential of rules and SVM feature selection evaluation of single features for ranking purposes.