The topic of interest in this study is herding behaviour in the stock market. We propose a new herd measure which is based on the cross-sectional dispersion of beta to detect the prevalence of herding phenomenon towards the market. The application of our new measure to TUNEDEX and BVMT indexes- as representatives of the Tunisian market- provide better results than those obtained by the cross-sectional stock priceís models developed by Christie and Huang (1995), Chang, Cheng and Khorana (2000) and Hwang and Salmon (2001, 2004). Moreover our findings show that the herd phenomenon consists of three essential components: constant term of herding which signals the existence of the phenomenon whatever the market conditions, the error term relative to the anticipations of the investors concerning the totality of assets, and the third component highlights that the current herding depends on the previous one.