Abstract State‐of‐the‐art models for soil compaction express soil strength by the so‐called precompression stress. The idea is to avoid stresses higher than those previously exposed to the soil. However, the quantification of soil precompression stress generally implies a problematic logarithmic transformation of the applied stress. In the present study, we analyzed strain‐stress curves from samples collected in the subsoil of a loess soil. Confined, uniaxial compression tests were performed with a strain‐controlled stress application to ∼800 kPa. We calculated soil compressibility as the increase in strain per unit increase in stress, that is, we treated strain and stress in linear scales. At low stresses, soil compressibility generally decreased with stress. For ∼88% of the samples, a local minimum in compressibility was observed and taken as a transition from strain hardening to strain softening. This, in turn, is considered analogous to a precompression stress. The median value for all tests was 75 kPa. For ∼12% of the samples, soil compressibility decreased in all the stress range. Further, 33 out of 58 samples with a local minimum in compressibility displayed yet another local minimum in compressibility at a higher level of stress. We strongly recommend the use of linear scale strain and stress in analyses of data from compression tests. Our study shows that soil will not necessarily possess a distinct precompression stress. Additional analyses in our study indicate that spline interpolation applied to data with few (∼11) observations provides reliable estimates of strain in between observations.