The last few years have seen a surge in drone popularity - not only in military and industrial applications, but also as for mainstream recreation uses. Despite its commercial success - or perhaps as a result of it - there are growing concerns on the risk they pose to aerial security as well as invasion of privacy. Detecting and classifying flying objects have always been a big research topic in the radar community. The challenge of detecting, tracking and classifying these small, sometimes multi-propelled flying objects is two fold. The first is the low radar cross section, which makes it hard for radars to pick up such a low energy echo. The second, once the echo has been detected, is classifying what kind of target it is (i.e bird, single propeller UAV or a multi-propeller drone). Tackling this challenge begins with giving the mathematical and physical model for the micro-Doppler (uDoppler) effect of a drone's radar returns. Then, this model will be used to train a multilayer per-ceptron (MLP) artificial neural network (ANN) to accurately classify a drone. Moreover, it will be shown that the MLP can be used also for regressing on the drone's propeller parameters such as blade length and frequency of rotation and to determine how many blades and propellers the drone consists of. All this will be attained from the baseband signal of the radar return. The work presented is essentially a proof-of-concept that radar-based complex target classification can be done effectively, while opening a new field of research of radar drones classification.