The surface roughness of Mars at scales from ~1--200 m is of interest for lander hazard and radar-sounder clutter analysis, but MOLA data are the only current source of global topographic information. We use synthetic fractal profiles to determine the uncertainties associated with deriving self-affine statistics from detrended segments of MOLA tracks. The mean Hurst exponent derived from such segments is an underestimate of the true value, H, for H > 0.5, and an overestimate for H 100-m scales. We suggest that any extrapolation be regarded as a lower bound on the true terrain statistics.