A method to resample the nonlinear least squares (NLS) estimator of the parameters of a regression model, which is nonlinear in the frequency domain, is proposed. The method uses the Fourier transform to remove the serial correlation that is present in the time domain. If the regressor series have seasonal variation or nonzero mean this creates additional problems for the resampling. A simple way to avoid these problems is proposed. The method is shown to give bootstrap replicates of the estimator that asymptotically have a distribution close to the true sampling distribution of the estimator. The proposed method is applied to a regression model that relates the inner and outer temperatures of a slab of insulation material to the heat flow through the slab