Normal scatter has been introduced into the free drug concentration variable, in sets of perfect binding data consisting of free (D f) and corresponding total (D t) drug concentrations. The standard deviation of error ranged from 0.1 × 10 −4 M to 1 × 10 −8 M over the free drug concentration range, 9 × 10 −4 M to 1 × 10 −8 M. These simulated data for 4 hypothetical drugs together with real equilibrium dialysis data for two non-steroidal antiinflammatory drugs, ibuprofen and naproxen, have been used to evaluate various non-linear regression procedures and commonly used weighting schemes, for calculating the binding constants from the Scatchard multiclass drug—protein binding model. It is statistically most desirable to regress unmanipulated primary experimental data, the independent variable being that measured most accurately. However, regression of unweighted D f on D t results in parameter estimates far removed from their true values. Appropriate data weighting is essential for this approach to be successful. Regression of either weighted or unweighted D f on the molar binding ratio, v̄, also produces poor estimates. It is shown that the computationally simpler regression of v̄ on D f performs as well as D f on D t (weighted 1/y), especially where good experimental data are obtained. This is of interest considering the greater computational sophistication required to regress D f on D t.