Objective: To assess the diagnostic performance of 3D sampling perfection with application-optimised contrasts using variable flip-angle evolution (SPACE) turbo spin-echo (TSE) sequences compared to 2D TSE for comprehensive knee assessment at 3 T. Methods: From January to July 2011, isotropic 3D SPACE was added to a 2D knee protocol at 3 T. Forty patients underwent subsequent arthroscopy. Three readers independently assessed MR images for meniscus, anterior cruciate ligament (ACL) and cartilage lesions. Readers 1 and 2 evaluated 3D and 2D data at separate sittings; reader 3 interpreted the complete exam including 3D and 2D sequences. Accuracies were calculated using arthroscopy as reference standard. McNemar's test (p < 0. 05) was used to compare 3D and 2D techniques. Results: The highest diagnostic yield was obtained by reader 3 (accuracies ≥88 %). For the medial meniscus, readers performed better with the 2D technique than with 3D SPACE (accuracies 85-88 % vs. 78-80 %, respectively) (p > 0. 05). For the lateral meniscus and ACL, 3D and 2D techniques had similar performance (accuracies ≥93 %). For cartilage lesions, 3D SPACE had significantly lower specificity (p = 0. 0156) than the 2D protocol for one reader. Conclusion: The conventional 2D TSE acquisition is more reliable than 3D SPACE for comprehensive assessment of the knee at 3. 0 T. Main Messages: • 3D SPACE is a valuable component of a knee MR protocol at 3 T.• 3D SPACE cannot be used as a single sequence in the MR evaluation of the knee at 3 T.• Knee MR protocols at 3 T should include both 2D and 3D TSE sequences.