Most sensory systems encode external signals into action potentials for transmission to the central nervous system, but little is known about the cost or efficiency of this encoding. We measured the information capacity at three stages of encoding in the neurons of a spider slit-sense mechanoreceptor organ. For the receptor current under voltage clamp, the capacity was approximately 1400 bits/s, but when the neuron was allowed to generate a receptor potential, nonlinear membrane processes improved the capacity to >2000 bits/s. Finally, when action potentials were produced, the capacity dropped to approximately 200 bits/s, or approximately 14% of the receptor current capacity. These measurements provide a quantitative estimation of the cost of encoding analog signals into action potentials.