This document proposes a video genre classification method using only audio features extracted from video clips. It uses Multivariate Adaptive Regression Splines (MARS) to build classification models for different genres based on low-level audio features like MFCCs, zero crossing rate, short-time energy, etc. extracted from a dataset of news, cartoons, sports, music and dahmas video clips. The models are able to accurately classify video genres with an overall classification rate of 91.83% based on the important audio features identified for each genre by the MARS models.