The early differentiation of Alzheimer's disease (AD) from frontotemporal dementia (FTD) may be difficult. The Tower of London (ToL), thought to assess executive functions such as planning and visuo-spatial working memory, could help in this purpose. Twentytwo Dementia Centers consecutively recruited patients with early FTD or AD. ToL performances of these groups were analyzed using both the conventional statistical approaches and the Artificial Neural Networks (ANNs) modelling. Ninety-four non aphasic FTD and 160 AD patients were recruited. ToL Accuracy Score (AS) significantly (p < 0.05) differentiated FTD from AD patients. However, the discriminant validity of AS checked by ROC curve analysis, yielded no significant results in terms of sensitivity and specificity (AUC 0.63). The performances of the 12 Success Subscores (SS) together with age, gender and schooling years were entered into advanced ANNs developed by Semeion Institute. The best ANNs were selected and submitted to ROC curves. The non-linear model was able to discriminate FTD from AD with an average AUC for 7 independent trials of 0.82. The use of hidden information contained in the different items of ToL and the non linear processing of the data through ANNs allows a high discrimination between FTD and AD in individual patients.