This paper separates the components of capital appreciation returns in an asset market into fixed and stochastic portions. It proposes a control for the problem of fixed components in the capital appreciation return used in transaction-based estimates. We find a consistent bias in the index resulting from repeat sales regressions which may be eliminated through simple methods. The sign and magnitude of the bias, as well as its systematic variation across properties, suggest that it is caused by incremental home improvements, as well as by price risk. We propose a maximum likelihood method or estimating the first and second moments of the fixed and temporal components of real estate returns that relies upon relatively small samples. In samples of 250 repeat-sales observations, it provides reliable estimates of the non-temporal and temporal return components, the variance of these components, and of the average correlation across individual properties in the market.