This paper describes the development of an inference system used for the identification of genes that encode enzymes of metabolic pathways. Input sequence alignment values are used to classify the best candidate genes for inclusion in a metabolic pathway map. The system workflow allows the user to provide feedback, which is stored in conjunction with analysed sequences for periodic retraining. The construction of the system involved the study of several different classifiers with various topologies, data sets and parameter normalisation data models. Experimental results show an excellent prediction capability with the classifiers trained with mixed data providing the best results.