1. By physical nature the critical brittleness temperature is a random variable and can be represented by the statistical distribution. 2. For steel 20L the type of fracture is a more accurate criterion for determining Tcri than the impact strength, and the standard deviation of % D increases sharply in the critical temperature range; near the limits of the critical temperature range the statistical distributions of % D change sharply, passing from bell-shaped to the J type. Also, the index of the type of fracture changes rapidly at the limits of the critical temperature range, changing from random to reliable or impossible, which does not occur in the case of the impact strength. 3. The critical temperature for steel 20L determined on the basis ofan=2 kg-m/cm2 is too low.