In this paper, we consider three techniques for radius estimation of subsurface metal pipes using Ground Penetrating Radar (GPR) data. The GPR acquisition system was Zond-12c with shielded antennae of frequency 900 MHz. Seven buried pipes from 0.01 m up to 0.36 m in radius were studied. For GPR data processing were applied the methods of weighted least squares, the recursive Kalman filter and the maximum likelihood. The recursive Kalman filter method shows the minimum value for average error of 3.08% for radius estimation. The maximum value for average error of 13.3% for all considered methods corresponds to pipe with the minimum radius of 0.01 m. For pipes with radius from 0.024 m up to 0.36 m the average errors are varied in the range of 0.74 - 5.17%. We studied the influences of random errors of measurements on the accuracy of radius estimations for the metal pipe of radius 0.36 m. We also inspected the influence of the angle between the antennae scan direction vector and the pipe axis on the accuracy of radius estimations by recursive Kalman filter technique for all studied pipes.