This paper deals with sideslip angle estimation of powered two-wheeled vehicles. Since available sensors used to directly measure this variable are bulky and expensive, estimation algorithms based on on-board measurements have been developed. These algorithms are mainly devoted to four-wheeled vehicles whereas the sideslip estimation for two-wheeled vehicles is still an open topic. This paper presents a Neural Network estimation algorithm that uses on-board standard measures available in modern motorbikes and studies the role of the most significant signals for the estimation. The employed black-box approach does not require the derivation of any physics-based model of the motorcycle dynamics and thus is meant as a valid tool for a preliminary insight in such estimation problem. The experimental data collected cover a rich amount of manoeuvres that are used to train the network and several other manoeuvres have been used to analyse its performances.