The performance of ground-based surveillance radars strongly depends on the distribution and spectral characteristics of ground clutter. To design signal processing algorithms that exploit the knowledge of clutter characteristics, a preliminary statistical analysis of ground-clutter data is necessary. We report the results of a statistical analysis of X-band ground-clutter data from the MIT Lincoln Laboratory Phase One program. Data non-Gaussianity of the in-phase and quadrature components was revealed, first by means of histogram and moments analysis, and then by means of a Gaussianity test based on cumulants of order higher than the second; to this purpose parametric autoregressive (AR) modeling of the clutter process was developed. The test is computationally attractive and has constant false alarm rate (CFAR). Incoherent analysis has also been carried out by checking the fitting to Rayleigh, Weibull, log-normal, and K-distribution models. Finally, a new modified Kolmogorov-Smirnoff (KS) goodness-of-fit test is proposed; this modified test guarantees good fitting in the distribution tails, which is of fundamental importance for a correct design of CFAR processors.