Abstract Soil compaction is a serious threat to soil functions and ecosystem services. Persistent soil deformation takes place when mechanical stress exceeds soil strength. Risk assessment models typically assume soil to be elastic to a precompression stress level and plastic above this threshold. Currently used procedures for estimating soil precompression stress imply applied stress in a logarithmic form that has been criticized. We performed uniaxial, confined compression tests on 117 soil samples with a well‐defined stress history. Strain and stress were recorded at 200 levels of stress in the range of 0–1 MPa. Soil compressibility was calculated as incremental strain per incremental stress, using stress in arithmetic scale. For a given sample, a local minimum in compressibility appeared close to the preload applied to the sample prior to the compression test. This yield stress is suggested as an expression of soil precompression stress. Not all samples displayed a yield stress. Primarily soil exposed to a high preload did not exhibit a clear stress level with change in compressibility. This indicates that a physically based stress level pointing out a transition to plastic conditions will not exist for all soils. Our observation calls for new concepts in risk assessment. Tests with interpolation between a limited number of data points indicate that the yield stress—if existing—may be detected from classical compression data.