This paper provides a general theoretical and methodological framework for the estimation of the translog production function using panel data. The estimation is carried out using generalised least squares estimator (random-effects model) under different maintained hypotheses for the underlying production technology. Our empirical results indicate that imposing a priori restrictions such as homotheticity, homogeneity or separability on the production structure are not palatable and should rather remain a testable hypothesis within the estimation framework.