A non‐parametric empirical approach, called the conditional average estimator (CAE) method, has been implemented for the estimation of the flexural deformation capacity of reinforced concrete rectangular columns expressed in terms of the ultimate (‘near collapse’) drift. Two databases (PEER and Fardis), which represent subsets of the original databases, were used. Four input parameters were employed in the basic model: axial load index, index related to confinement, shear span index, and concrete compressive strength. The results of analyses suggest that, in general, ultimate drift decreases with increasing axial load index, and increases with better confinement. An increase in the shear span‐to‐depth ratio has a beneficial effect until a turning point is reached. After that the opposite trend can be observed, i.e. a decrease in the ultimate drift with further increasing of the shear span‐to‐depth ratio. No clear trend is observed in the case of concrete compressive strength. The predictions, obtained by using the Fardis database are in general somewhat larger than the predictions from the PEER database, due to the difference in the definition of ultimate drift. The scatter of results is large. The local coefficient of variation, which is a measure for dispersion, amounts to about 0.2–0.5. The ultimate drifts obtained by using the two databases, were compared with the values predicted by the Eurocode 8 empirical formula. Copyright © 2006 John Wiley & Sons, Ltd.