A central problem in estimating per unit costs of production originates from the fact that most farms produce multiple outputs and standard farm-accounting data are only available at the whole-farm level. The seemingly unrelated regression (SUR) approach is used to estimate per unit production costs based on German farm accountancy data. Special emphasis is put on outlier detection prior to the estimation of production costs to increase the robustness of the results. Outlier observations are identified based on the Mahalanobis distance for each observation on the data set. It was observed that less negative cost coefficients are estimated after the exclusion of the outliers. The time series analysis of cost estimation based on SUR regression shows the costs of arable crops after 2004, affected by rising prices of fertilizer, seeds and energy, while the increase of livestock production costs after 2006 is attributed to feed costs.