Due to the heterogeneity of soil, there are spectral differences between different measurement positions of a soil sample and among different samples of a soil genetic layer. As a result, an estimated value of a soil property using in situ spectra of soil inevitably has errors. However, these errors have not been revealed so far. In this paper, 80 soil profiles and 38 soil surface samples were collected twice with cutting ring from a small area of woodland in typical hilly areas in southern China. Then, the soil organic matter content was measured by in-situ spectrometry and traditional laboratory method, so as to analyze the estimation error of soil organic matter content caused by different spectral test points and sampling locations. The results show that, the spectral difference of each sample at a total of 18 test points ranges from 0.12° to 8.13°, with an average value of 1.55°. The spectral difference between two repeated sampling locations of each sample is 0.18°~3.65°, with an average value of 0.88°. The estimated error of soil organic matter due to the different positions of test points was 0.92~14.66 g·kg^-1, accounting for 3.8%~428% of the measured organic matter content. The estimation error of soil organic matter caused by different sampling locations is 0.005 7~11.46 g·kg^-1, accounting for 0.017%~92% of the measured organic matter. Moreover, the error caused by the former is larger than that caused by the partial least squares regression model used in this paper, while the error caused by the latter is slightly smaller than that caused by this model. In addition, it is found that these two errors increase with the increase of measured organic matter content. Hence, this paper argues that the errors caused by different test points and sampling locations should be paid attention to in future studies, especially in soils with higher organic matter content, and calls for research on more effective methods to reduce these errors.