Variable angle fixed arm peel and mandrel peel tests were performed on four metal-polymer laminate systems. In total, four polymeric adhesives and three grades of aluminium alloy (AA) substrates were used, enabling a wide range of material properties to be encompassed in the study. Mandrel peel tests provided a direct determination of the plastic bending energy (G(p)) and adhesive fracture toughness (G(a)). For the fixed arm tests, a global energy-balance analysis (ICPeel software) was used to determine G(a) and G(p) analytically. This was done via the calculation of the maximum curvature of the peel arm (1/R-0) and the root rotation angle (theta(0)) from a beam on elastic foundation model. In order to investigate the accuracy of the analytical approach, an experimental method based on high resolution digital photography enabled 1/R-0 and theta(0) to be measured independently. It was then possible to compare these parameters by measurement and by analytical approach (ICPeel software). theta(0) and R-0 relate to the slope and curvature of the peel arm at the debonding front, respectively. In order to measure these parameters, the coordinates of the edge of the peel arm were extracted from each digital photograph, and the slope and curvature were calculated numerically from these curves. The crack tip was then defined as the point of maximum curvature 1/R-0, in accordance with traditional beam theory. It was found that the smoothing in the calculation of first and second derivatives could generate significant errors in the value of theta(0). On the other hand, R-0 was found to be a more robust measurement, with little dependence on smoothing. Nevertheless, on most occasions, the measured values of theta 0 and R-0, as well as the resulting G(a) were shown to be in good agreement with the analytical model. Since the peel fractures were generally cohesive, G(a) was compared with the cohesive fracture toughness (G(c))obtained from Tapered Double Cantilever Beam (TDCB) tests with a fracture mechanics analysis. Good agreement was observed, confirming that G(a) is likely to be a geometry-independent fracture parameter. (c) 2005 Springer Science + Business Media, Inc.