When subjected to elevated temperatures, steel displays a reduction in both strength and stiffness, its yield plateau vanishes and its response becomes increasingly nonlinear with pronounced strain hardening. For steel sections subjected to compressive stresses, the extent to which strain hardening can be exploited (i.e. the strain at which failure occurs) depends on the susceptibility to local buckling. This is reflected in the European guidance for structural fire design EN1993-1-2 [1], which specifies different effective yield strengths for different cross-section classes. Given the continuous rounded nature of the stress–strain curve of structural steel at elevated temperatures, this approach seems overly simplistic and improved accuracy can be obtained if strain-based approaches are employed [2]. Similar observations have been previously made for structural stainless steel design at ambient temperatures and the continuous strength method (CSM) was developed as a rational means to exploit strain hardening at room temperature. This paper extends the CSM to the structural fire design of steel cross-sections. The accuracy of the method is verified by comparing the ultimate capacity predictions with test results extracted from the literature. It is shown that the CSM offers more accurate ultimate capacity predictions than current design methods throughout the full temperature range that steel structures are likely to be exposed to during a fire. Moreover due to its strain-based nature, the proposed methodology can readily account for the effect of restrained thermal expansion on the structural response at cross-sectional level.