Among the many controversial variables in finance, risk premia stand prominent for their lack of observability. Measuring premia as the difference between realized returns on risky and risk-free assets has not led to unanimous conclusions about their size, which dramatically depends upon the length of the sample; in addition, investment allocations or inflation expectations are influenced by ex-ante values of the risk premia and ex-post returns are, if any, rough approximations of these. A countless number of papers have dealt with this issue, from the initial contribution of Mehra and Prescott (1985) to very recent advances within a bayesian framework of Pastor and Stambaugh (2001). This paper uses univariate and bivariate conditional variance models as approximations of static and intertemporal Capm schemes; the size of the equity premium is assessed for the US both at the market level and, through a conditional version of the three-factor model of Fama and French (1993), at a firm-level. The market premium has had large swings with short-lived peaks over the last 75 years, floating around a mean value of 5 per cent on a yearly basis; this value rises to 6.5 percent when time-varying investment opportunities are allowed for. In periods of economic expansion the expected premium on the equity return is nearly half the value expected in recession, twenty percent less if the Great Depression period is excluded; the cross-sectional dispersion of the firm-level premia as a function of firm's size is also influenced by the position of the economy within the business cycle.