While the risk premium hypothesis in futures markets has been the subject of a long and continuous controversy, the risk premium hypothesis in forward markets is also of interest among economists. The hypothesis is supported by some theoretical arguments and empirical evidence yet remains an open question. We in this study apply a two-equation regression model similar to those used in (Fama and French (1987} and de Roon et al. (1998) to analyze the risk premiums in forward markets, particularly, using the pre-harvest wheat forward markets in Illinois (1982-2004) and Kansas (1990-2004) as an example. The two-equation regression model consists of a forecasting equation, which uses a forward basis during a pre-harvest period to forecast the spot basis at the following harvest period, and a risk premium equation, which uses the forward basis to predict the risk premium to be realized at the harvest. The empirical results show that, first, the average realized risk premiums for Illinois fluctuate around a level during the entirety of a pre-harvest period, while the risk premiums for Kansas show a slight downward trend as time approaches the harvest. The average realized risk premiums are generally positive and bigger for Illinois than for Kansas, but all mean risk premiums are within one units of their corresponding standard deviations. Second, the pre-harvest forward bases have reliable forecasting power for the spot harvest bases and contain information regarding the risk premiums, which strongly recommend estimating risk premiums conditional on forward bases.