Abstract It has been proposed that grade A should be replaced with a tougher steel for ship construction. This recommendation is not supported by casualty statistics. Brittle fractures have decreased from once every 10 ship years at the time of the Liberty ships, to once every 100 ship years in the 1950s, and now stand at around once every 10,000 ship years. The authors have developed a probabilistic fracture mechanics approach indexed to the Charpy 27 Joule temperature, T 27J. This is used here to analyse the Liberty ship failures. Comparison with a databank of T 27J values for grade A steel spanning 1920 to 1995 suggests that the Liberty ship steel was unusually brittle. The probability analysis shows good agreement between predicted and observed fracture casualty rates for the Liberty ships and helps explain why brittle fractures are now uncommon in grade A steel ships.