<p>This article presents a new statistical method for detecting defects in timber I-beams using normality tests of deflection line linear approximation error distribution. The method uses static deflection measurements obtained through a digital image correlation (DIC) system. It relies on the observation that normality in the error distribution is lost more quickly with the extension of the approximation sections in areas where the beam’s curvature increases, including causes related to material defects that reduce flexural stiffness. A key advantage of this approach is that it identifies the areas with increased curvature using a defined level of statistical significance when rejecting the null hypothesis in normality tests. So far, this type of damage detection in structural beams has not been explored, which was a motivation to conduct this research. The method was tested using both theoretical models and experiments on timber composite beams with softwood flanges and an oriented strand board web. It successfully identified both artificially introduced damage and natural stiffness reductions caused by knots in the tension flange.</p> Graphical abstract <p></p>