Strength characteristics of steel beam-columns are discussed by statistically evaluating the data obtained from previous experimental studies. A total of 237 test data are selected in this study, and their individual properties like material and geometrical properties and maximum resistances are stored in a computer in the format of a relational data base. Strength of the test data (designated as the experimental strength) is calibrated against the strength assumed in the strength provisions of the LSD Specification (designated as the predicted strength), and following observations are made. (1) The experimental strength is 15 % larger on the average than the predicted strength. (2) The experimental strength relative to the predicted strength is larger for beam-columns with moment-gradient than for those with uniform-moment, and it occurs most likely because the effect of strain hardening on the strength is not included in the predicted strength. (3) If the nominal yield stress is used instead of the real yield stress, the difference between the experimental and predicted strengths is expanded further by 7 %. (4) The experimental strength is 36 % larger on the average than the design strength stipulated in the LSD Specification. (5) The primary source that have produced the difference between the experimental and design strengths is not the variation of yield stress but the error involved in the predicted strength.