This work presents a simple and fast artificial Neural Network (NN) with an hierarchical feed-forward topology and limited number of neurons, which classifies proteins into two classes, from their sequences alone: the membrane protein class and the non-membrane one. By using only information contained in 11 protein sequences, the method was able to identify with 100% accuracy 155 membrane proteins with reliable topologies collected from several papers in the literature. Applied to a test set of 995 globular, water-soluble proteins, the NN classifies correctly 98% of them. The method was also applied to the whole SwissProt database with 94% success and on ORF's of several complete genomes. We also developed a new method (PRED-TMR) that predicts the location of transmembrane (TM) domains in proteins from their sequence alone. Α standard hydrophobicity analysis is refined with the detection of potential edges of TM regions. This allows both to discard highly hydrophobic regions not delimited by clear start and end configurations and to confirm putative TM segments not distinguishable by their hydrophobic composition. The accuracy of PRED-TMR obtained on a test set of 101 non homologous TM proteins with reliable topologies compares well with that of other popular existing methods. Prediction accuracy slightly decreased when PRED_TMR was applied to all TM proteins of SwissProt. The two methods were integrated in a package (PRED-TMR2), which can be used for the functional assignment and analysis of newly sequenced genomes and, especially, those ORF\textquoteright\s that correspond to proteins with unknown function. A WWW server running PRED-TMR2 is available at http://biophysics.biol.uoa.gr/PRED-TMR2.