This paper estimates a dynamic microeconometric model of housing supply. The model features forward-looking landowners who optimally choose both the timing and the nature of construction while taking into account expectations about future prices and costs. The model is estimated using a unique dataset describing individual landowners in the San Francisco Bay Area. Results indicate that geographic and time-series variation in costs are key to understanding where and when construction occurs. Pro-cyclical costs provide an incentive for some landowners to build before price peaks. Results also indicate that landowners actively "time" the market, which reduces the elasticity of supply.