Eight models are examined as input-output representations of steady-state vision in humans at moderate to low level illumination. Three new models for visual contrast sensitivity are introduced and evaluated using contrast sensitivity function (CSF) data with samples on both narrow and wide frequency ranges. Additionally, five variations of previously published models are evaluated using the same data. A nonlinear least squares fitting algorithm produced the optimal parameters for each model. The eight models are compared on the basis of RMS error in their fit to the CSF data. The three new models, based on second-, third-, and fourth-order filter functions, provided the best fit to the data. They appear to more-closely approximate the underlying sensory mechanisms, and thus they provide a more useful input-output representation of the overall human visual system. >