Average life expectancy has increased steadily in recent decades. This phenomenon, considered together with aging of the population, will inevitably produce in the next years deep social changes that lead to the need of innovative services for elderly people, focused to improve the wellbeing and the quality of life. In this context many potential applications would benefit from the ability of automat-ically recognize facial expression with the purpose to reflect the mood, the emotions and also mental activities of an observed subject. Although facial expression recog-nition (FER) is widely investigated by many recent scientific works, it still remains a challenging task for a number of important factors among which one of the most discriminating is the age. In the present work an optimized Convolutional Neural Network (CNN) architecture is proposed and evaluated on two benchmark datasets (FACES and Lifespan) containing expressions performed also by aging adults. As baseline, and with the aim of making a comparison, two traditional machine learning approaches based on handcrafted features extraction process are evaluated on the same datasets. Experimentation confirms the efficiency of the proposed CNN archi-tecture with an average recognition rate higher than 93.6% for expressions performed by ageing adults when a proper set of CNN parameters was used. Moreover, the exper-imentation stage showed that the deep learning approach significantly improves the baseline approaches considered, and the most noticeable improvement was obtained when considering facial expressions of ageing adults.