Recent evidence by Fama and French (1992,1996) and others shows that betas and returns are not related empirically. They interpret this as evidence against the validity of the capital asset pricing model and they conclude that the beta is not a good measure of risk. This paper claims that usual tests do not leave much opportunity for beta to appear as a useful variable capable of explaining returns, because tests are often performed in periods where the average realised market excess return is not significantly different from zero. In order to assess the usefulness of beta, I propose an alternative approach that dissociates results obtained in period where the realised market excess is positive from those where it is negative. These new tests are then applied to a representative sample of the Swiss stock market over the period 1983-1991. The different results unambiguously support the fact that beta is a good measure of risk, because beta is strongly related to the cross section of realised returns. These results also confirm that there are no arbitrage opportunities on this market