Determination of soil precompression stress (σp) from compression curves has markedly changed in recent decades, with several available methods of varying complexity. The graphical procedure proposed more than eight decades ago remains a standard, though it is now emulated by statistical and numerical procedures. We tested nine determination methods based on linear, polynomial, nonlinear and spline regressions. Four attributes were used to represent soil deformation during compression (strain-attributes): bulk density (ρb), void ratio (e), total porosity (ϕ) and strain (ε). Undisturbed soil samples equilibrated either at 10 or 100 kPa water tension were subjected to confined, drained uniaxial compression tests. The experimental factors (determination method, strain-attribute, and water tension) and their interaction affected the determination of σp. Sigmoid equations converged for 84%–95% of the compression curves, failing for samples at 100 kPa with with steep virgin compression lines. Methods based on fourth-degree polynomials resulted in 10-94% of the values outside the range 10–1550 kPa, depending on the strain-attribute. Although ε and e are linearly related, the former suits better the application of methods based on polynomial and spline regression, also fitting well for methods based on sigmoid equations with samples at 10 kPa. The bi-linear intersection methods presented relaistic σp values for bi-linear curves and should be solely applied to ρb. The spline method resulted in σp values clustered along the loads applied during the compression tests.