White noise analysis methods for characterizing neurons typically ignore the dynamics of neural spike generation, assuming that spikes arise from an inhomogeneous Poisson process. We show that when spikes arise from a leaky integrate-and-fire mechanism, a classical white-noise estimate of a neuron's temporal receptive field is significantly biased. We develop a modified estimator for linear characterization of such neurons, and demonstrate its effectiveness in simulation. Finally, we apply it to physiological data and show that spiking dynamics may account for changes observed in the receptive fields measured at different contrasts.