This dissertation introduces an innovative technique to extract ground elevation models using smallootprint LIDAR data. This technique consists of a preprocessing step, ground modeling, and interpolation. In the preprocessing step, much of the non-terrain points are eliminated using a histogram-based clustering technique. Then, in the ground modeling stage, the information such as elevation and slope between nearest neighbor points is extracted. This step corresponds to an outlier detection process. In this stage, residuals and gradient indices for elevation and slope, are introduced. These indices are investigated for a constructed 95% confidence interval to discard the remaining non-terrain points. Finally, using spline interpolation, a smooth ground surface is generated. Experimental results show that the presented technique is more robust and yields better results compared to existing techniques, such as linear prediction, modified linear prediction, and adaptive smoothing, in terms of the root mean squared error, absolute mean, and absolute standard deviation. Furthermore, the possibility of using a ground trend model developed from specific tree height measurements is investigated. For this analysis, a statistical regression analysis model is used. Performing this analysis, a 0.63 R-squared value is obtained. This result indicates that the LIDAR ground surface obtained from the presented algorithm is related to the true ground surface.