[en] Specimen size effects on K_Jc data scatter in the transition range of fracture toughness have been explained by external (weakest link) statistics. In this investigation, compact specimens of A 533 grade B steel were tested in sizes ranging from 1/2TC(T) to 4TC(T) with sufficient replication to obtain good three-parameter Weibull characterization of data distributions. The optimum fitting parameters for an assumed Weibull slope of 4 were calculated. External statistics analysis was applied to the 1/2TC(T) data to predict median K_Jc values for 1TC(T), 2TC(T), and 4TC(T) specimens. The distributions from experimentally developed 1TC(T), 2TC(T), and 4TC(T) data tended to confirm the predictions. However, the extremal prediction model does not work well at lower-shelf toughness. At -150 degree C the extremal model predicts a specimen size effect where in reality there is no size effect