Estimation models of drug demand should encompass the aspect of addiction. Here, we consider two static panel data regression models and two cross-section models with lags or leads in drug consumption as additional regressors. Heroin injectors attending a needle exchange service in Oslo were interviewed twice, with a one-year interval. Despite our relatively small sample, we obtain statistically significant price and income responses for nearly all of the models and specifications applied. The sample is split by dealing status, with dealers obtaining price elasticities in the range of [−0.15, −1.51] and non-dealers [−0.71, −1.69]. Somewhat surprisingly, the estimates of the variance of the latent individual-specific variable are rather low in the panel data models, although higher for non-dealers than for dealers.