It is argued here that in the presence of joint production, ordinary least squares regression, OLS, gives inconsistent estimates. An alternative estimation procedure is therefore developed. The application of the powerful tool of canonical correlation analysis for econometric problems has so far been rather limited because of certain problems. An attempt is made in the first part of this paper to extend the analysis in order to handle a greater variety of problems. A New way of looking at the canonical variables is suggested while developing an estimation technique for joint production functions. In the second part, empirical examples and a comparison with OLS are included. The forecasting accuracy is also compared, illustrating the superiority of the suggested estimation procedure.