This study deals with modeling of production risk by means of a two-step procedure. In contrast to earlier studies of production risk, we do not immediately adopt restrictive functional forms for the risky production technology. We first test for the presence of production risk. If production risk is found to be present, the mean and risk functions are estimated separately. This allows the use of more flexible functional forms for both the mean and the risk functions than commonly found in the literature. An empirical application to Norwegian salmon farming, where restrictive specifications of the technology are rejected, demonstrates the validity of our approach. Presence of production risk many primary production sectors implies that this approach should be considered in productivity studies.