An anomoly exists in the literature of radiation-induced chromosomal aberrations. While X- or gamma-ray induced chromosomal exchanges observed at metaphase increase approximately as the square of the dose, anaphase bridges that are supposedly the descendants of these metaphase exchanges increase only about linearly with dose. Metaphase exchanges (dicentrics and rings) could fail to produce observed anaphase bridges if (a) the exchanges separate in a fall-free (non-bridge-producing) manner, or (b) the bridges that are produced at separation rupture before observation at anaphase. By detailed observation in Tradescantia of dose curves for metaphase exchanges and anaphase bridges, and also of exchanges that separate in the fall-free (no bridge is formed) or interlocked (bridge-producing) manner at “separation” stage (very earliest anaphase), it was possible to resolve the problem. A constant fraction, 47 per cent independent of dose, of metaphase exchanges separate in the fall-free manner, 53 per cent separate to make bridges. But of the bridges formed at separation, a fraction, increasing exponentially with dose, rupture at early anaphase and thus escape detection as anaphase bridges in the usual scoring procedure. About 19 per cent of the bridges rupture at 100 r, and this fraction increases exponentially to about 72 per cent at 400 r. A model is proposed to explain rupture of bridges in anaphase, based on the number of restituted breaks (or potential or partial breaks), proportional to dose, in a bridge. It is assumed a restituted break (or potential or partial break) constitutes a weak spot in the chromosome that has the possibility, P b , of rupturing under tension in an anaphase bridge. A bridge containing n such restituted breaks in it ( n proportional to dose) has a probability of rupture, P r = 1 − (1 − P b) n , thus the probability of rupture increases exponentially from zero at zero dose to 1·0 at large dose. The model predicts that at very high dose (at very high restituted break frequencies) observed anaphase bridges should fall to zero, and that with linear-with-dose exchange ancestors—as are formed with high LET radiations—anaphase bridges, as observed, should increase less rapidly than linear with dose. The model also predicts that the curve for anaphase bridges vs. dose should increase (not necessarily linearly) with dose, pass through a maximum and fall toward zero at higher doses. A large fraction, 33 per cent, of all rings separate to form a large single (dicentric) ring at anaphase. Obtaining the large ring at anaphase is most likely if the G 1 chromosome is single stranded and non polarized, less if functionally double-stranded and non polarized, and least likely if both two-stranded and polarized. Furthermore, the two-stranded polarized condition would produce particular configurational types at separation which are not observed. From the data of this experiment, it is calculated that scoring of “abnormal anaphases”, i.e. cells with bridges, detects roughly 20 per cent of the chromosomal damage actually produced and analyzable at metaphase. This value is not constant, being more at low dose, less at higher doses.