Two classes of models for pricing interest rate contingent claims are compared. The first contains standard single factor models, while the second includes stochastic volatility variants of these. It is demonstrated that incorporating stochastic volatility significantly improves the ability of the models to fit movements in short term interest rates. The fundamental reason for this is that standard univariate models do not generate enough conditional heteroskedasticity. The models are also compared on the basis of bond option prices when each is constrained to fit a particular yield curve. The results suggest that stochastic volatility models will typically produce lower option values than corresponding single factor models, particularly for near-the-money to deep-out-of-the-money options. This is because estimated historical volatilities are significantly lower under stochastic volatility. When single factor models are constrained to the same level of volatility as implied by stochastic volatility models, in most cases the pricing differences become minor.