Practical models for the ultimate deformations of flexure-controlled members were fitted to a database of thousands of tests. They cover beams, rectangular columns or walls, members with T-, H-, C-, or box section, conforming or not to seismic design codes, with or without fiber reinforced polymer jackets, in monotonic or cyclic uniaxial flexure with axial force. Model parameters were statistically derived assigning to each test (a) a weight of unity; (b) weight inversely proportional to the number of tests in a testing campaign, or (c) weight inversely proportional to the root-meansquare error of tests in a campaign. Predictions correlate well to those of a physical model based on curvatures and a plastic hinge length. Analysis of variance suggests that the scatter of model predictions with respect to test results comes more from the testing itself than from the model. Intralab and interlab variabilities are quantified and pure error is estimated from tests of nominally identical specimens.