It is difficult to exactly detect small objects in complex shallow subsurface. As the energy of backscattered signal from the target is low, backscattered signals from target and ground surface are overlapped. In addition, clutters created by ground surface and movement strongly depress the signaltoclutter ratio when shallowly buried small objects are detected by portable ground penetrating radar (GPR). An algorithm for detection of shallowly buried objects based on principal component analysis (PCA) is thus proposed in this paper. Via PCA decomposition, current Ascan data are projected onto the projecting direction of background data. A set test function is compared with adaptive threshold to decide if current Ascan data are from an object. Detection of shallowly buried small objects can be achieved in combination with background data dynamic updating. The data tested in sand, laterite, clay and lawn were processed, and the results show that shallowly buried objects can be detected using algorithm based on PCA.