The ocean-scattered Global Positioning System (GPS) signal can be used, in a bistatic radar configuration, for estimating the ocean surface roughness. This requires fitting a scattering model to the distribution of reflected power in delay and Doppler. The delay-Doppler map, or waveform, is generated through the cross-correlation of a local copy of the pseudorandom noise (PRN) code assigned to each GPS satellite with the reflected signal over increments in Doppler frequency equal to the pre-detection bandwidth. The correlation time of the reflected signal voltage sets an upper limit on the pre-detection integration time. Correlation time was estimated from experimental data through fitting a model Gaussian function to the magnitude of the complex autocorrelation of the time series of the waveform. Range bins were taken 1.1 chips prior to the averaged waveform peak in the leading edge and 2.5 chips after the averaged waveform peak in the trailing edge. Doppler bins were set at -500, 0, and +500 Hz relative to the Doppler frequency of the corresponding direct GPS signal. The direct GPS signal was tracked using a frequency locked loop (FLL). Results showed that maximum correlation time occurs approximately 0.72 code chips prior to the averaged waveform peak. The correlation time was found to vary with the elevation angle of the satellite, with lower elevation satellites showing longer correlation time. Some experimental results were compared with model predictions. A good comparison was usually found for delay bins near the specular point and both the model and experiment showed a decrease in the correlation time for longer delays. In the higher range bins the model tended to predict longer correlation times than those observed in the experimental data