Failure of PZT materials is governed by the crack resistance curve (R-curve). The R-curve was evaluated for a soft PZT: (a) in controlled fracture tests with single-edge-notched bending bars via an improved compliance method combining mechanical compliance and microscopic crack length measurement, (b) by fracture mechanics evaluation of tensile strength of Knoop-damaged specimens and (c) by completely stable crack extension tests with a loading device consisting of two pairs of opposite line loads. It was found that the R-curves obtained with different test methods differ strongly. A possible interpretation of the differences is given. A theoretical part deals with the determination of the stress intensity factor solution for bending bars with edge cracks as used in the experiments. Piezoelectric materials exhibit a non-linear stress-strain curve and non-symmetry in tension and compression. Under these conditions the non-linear stress distribution is computed for the bending bar and the stress intensity factor is determined by using the fracture mechanics weight function method. From these computations it results that maximum deviations from the linear-elastic solution of less than 2% occur if a/W > 0.3 (a= crack length, W= specimen width). In case of the roller loading, it can be shown that maximum errors must be less than 7%.