This paper shows that neglecting environmental range dependence in matched-field inversion (MFI) results in biased theory errors that lead to biased geoacoustic parameter estimates using standard inversion methods. Two approaches are used to investigate this issue. The first considers the distribution of optimal parameter estimates obtained from a large number of range-independent inversions of synthetic data generated for random range-dependent environments. The second applies Bayesian inversion and computes marginal uncertainty distributions for geoacoustic parameters, neglecting environmental range dependence. Both hard- and soft-bottom environments are considered at a number of scales of lateral variability for water depth and seabed sound speed. While the use of multifrequency data reduces the variability of the parameter estimates, it does not generally reduce parameter biases and increases biases in some cases. The biases appear to result from additional losses in range-dependent propagation, which are compensated for in range-independent inversion by adjusting geoacoustic parameters to decrease the seabed reflection coefficient. The effects of range dependence differ for different environments, with the soft-bottom case sensitive to range-dependent sound speed and the hard-bottom case sensitive to range-dependent water depth.