We use a semiparametric estimator to analyze the relationship between land values and parcel size in a sample of 158 undeveloped parcels in the Portland, Oregon, metropolitan area. The semiparametric estimator combines the benefits of parametric and nonparametric estimation. The value-size relationship is estimated nonparametrically, which permits the function to be linear, convex, and concave in different regions. A simple log-linear parametric relationship is assumed for the rest of the model, which conserves degrees of freedom and simplifies hypothesis testing. Our semiparametric estimates do not reject log-linearity for the value-size relationship.