In order to estimate the effects of unions on worker productivity, a Cobb-Douglas production function is modified so that unionization is included as a variable. The resulting functional form is similar to that used to isolate the effect of worker quality in previous studies. Using state by two-digit SIC observations for U.S. manufacturing, unionization is found to have a substantial positive effect on output per worker. However, this result depends on two important assumptions which we cannot verify directly; attempts to relax these assumptions are not conclusive.