Thirty-four anaestheized mongrel dogs subjected to thoracotomy were used to study AV node conduction during atrial pacing at increasing rates. Epicardial atrial electrodes were used, together with endocavitary recordings of His bundle electrogram. An analysis was made of the repercussions involved in using four different mathematical functions describing nodal conduction: three were nonlinear (exponential and hyperbolic A and B) and one linear. In the case of the first three, the consequences of using a direct nonlinear data-fitting procedure or an indirect procedure by linear transformations of the functions were studied. The exponential and hyperbolic B functions provided the least mean squared residual in quantifying nodal conduction (8.6 +/- 10.8 ms2 and 10.8 +/- 13.9 ms2, respectively). The use of nonlinear function transformation into a linear representation caused loss of precision in the fit to the data in the case of the exponential function (18.3 +/- 22.2 ms2 versus 8.6 +/- 10.8 ms2, p less than 0.01) and, to a lesser extent, in the case of the hyperbolic B function (12.5 +/- 16.4 ms2 versus 10.8 +/- 13.9 ms2, P less than 0.05).