Standard nonlinear regression techniques were used to fit nine data sets (21 points, cv = 0.001), derived from models that have been used to describe pesticide disappearance from soil, to each of the nine models. Goodness‐of‐fit was assessed using the residual mean square. Four models (3 four parameter and 1 five parameter) were found to be identical in their fit to each of the nine data sets. Additional data sets (21 and 8 points with cv's of 0.001, 0.01, 0.1 and 0.2) were generated for the remaining five models (1 four parameter, 1 three parameter and 3 two parameter) to evaluate the effect of number of observations and noise level on the data‐to‐model identifiability and the parameter estimates. The number of 21 point data sets uniquely identifiable with the model used to produce the data were 5, 3, 2 and 0 at cv's of 0.001, 0.01, 0.1 and 0.2 respectively. Similar results were observed for the 8 point data sets. At least one parameter was poorly defined in the three and four parameter models by the 21 point data set and in all five models by the 8 point data set with cv = 0.1. Five sets of observations for which one of the nine models was originally proposed were fitted to each of the five selected models, in addition to the original model. In all cases, simpler or previously described models fit the data as well as or better than the models proposed by the authors. At least one parameter was poorly estimated in most cases. Twenty‐six of the 45 sets of observations evaluated in the description of a recently proposed model based on spatial variability were fitted to each of the five selected models. In 17 cases, two parameter models fit the data as well as or better than the three parameter model proposed. The implications of these observations on the role of modeling in the study of pesticide disappearance from soil is discussed.