This study focuses on the spatial resolution required for cartilage imaging. The purposes of this study were (I) to analyze the diagnostic performance in diagnosing artificially produced cartilage lesions in a small joint model using an optimized fat saturated three-dimensional gradient-echo sequence, (II) to relate the lesion size and depth as diagnosed in the magnetic resonance images with the corresponding pathologic findings and (III) to assess signal-to-noise (SNR) ratios for each of the protocols. Twenty-five artificial cartilage lesions were created in the knee joints of 10 rabbits. These specimens and seven specimens without lesions were imaged at 1.5 T using a three-dimensional gradient-echo sequence with varying slice thickness, field of view and matrix. A total of 404 corresponding images were selected, 50% with and 50% without cartilage lesions. Six radiologists scored all images according to five levels of confidence and receiver operating characteristic (ROC) analysis was performed. Lesion size and depth were compared to the corresponding pathological specimen sections. Additionally SNR ratios were calculated. ROC analysis of pooled data from all readers showed the highest area under the ROC curve for the sequence with the highest spatial resolution, while the diagnostic performance was significantly lower in the other sequences (p <0.01). Assessment of the lesion size and depth was correct in 45% and 40% respectively with the highest resolution and in 29% and 23% with the lowest resolution. SNR ratios decreased with increasing spatial resolution. In conclusion this study shows that increasing spatial resolution improves diagnostic performance in cartilage lesions, though SNR decreases substantially. Assessment of correct lesion size and depth still is limited.