This study includes an estimate of cost functions for paddy farms in Formosa. Two types of estimational techniques are used; namely regression estimates and synthesized or budgeted cost curves. Long-run and short-run regression-derived curves are estimated for rice and sweet potato farms from samples which provide the observations. Estimates are for total cost functions with average cost functions computed per hectare. From sample observations, it was possible to derive, through regression estimates, only linear total cost functions. The non-linear average cost functions have only slight (economically unimportant) slopes beyond 2-5 hectares. Hence, large farms do not seem to be an important policy imperative. However, the steep slope of the average cost functions for farms under 1-5 hectares suggest that further farm fragmentation should be prevented. The cost estimates based on synthesized functions result in conclusions paralleling those based on regression-estimated cost functions. The area at which minimum average costs occur differs among sets of fixed inputs. However, the level of costs at minimum points on the curves are similar for the numerous resource situations examined. Within the limitations of the data and results, we believe both estimational techniques imply similar policy prescriptions.