<p>Tempered martensite embrittlement (TME) remains a major challenge for steels with lath martensitic microstructures because tempering can severely reduce impact toughness. This study compares two contrasting lath martensitic steels, a 0.33C-1.8Si steel and a 9Cr steel, to clarify how retained austenite decomposition, carbide evolution, and martensitic hierarchy govern brittle fracture under TME conditions. Hardness, Charpy impact toughness, and retained austenite volume fraction were measured as a function of tempering temperature, and the embrittled conditions were further characterized by SEM, EBSD, and TEM. A distinct toughness trough was found at 400&#xa0;°C in the 0.33C-1.8Si steel and at 500&#xa0;°C in the 9Cr steel. In both steels, embrittlement developed after a substantial reduction in retained austenite fraction, but did not coincide with the hardness peak. The 9Cr steel showed a much deeper toughness loss, associated with a coarser packet/prior-austenite structure and with the presence of coarse M₆C carbides and elongated cementite, whereas the 0.33C-1.8Si steel contained mainly fine intralath η-Fe₂C. In both steels, the mean quasi-cleavage facet size was intermediate between the mean packet size and the prior-austenite grain size. EBSD-based crack-path analysis revealed that in the 0.33C-1.8Si steel, cracks could partly follow prior-austenite grain boundary segments before switching to transgranular propagation, while in the 9Cr steel, cracks propagated mainly within prior-austenite grains and were deflected at packet boundaries. In both cases, transgranular crack traces were consistent with cleavage along planes close to {100}α. These results provide a unified view of TME across two different martensitic steel classes and show that the effective fracture unit is not a single block or packet, but a crack-path segment defined jointly by crystallographic compatibility and the local interfacial state produced by retained austenite loss and boundary carbide evolution.</p>