Objective To estimate the sample size of radiological diagnostic tests in multireader multicase (MRMC) design, and to explore the numbers of readers and cases under 3 different inference situations: random-reader random-case, fixed-reader random-case, and random-reader fixed-case. Methods The images of 114 participants (45 cases diagnosed as aortic coarctation by the gold standard) in the Van Dyke dataset were used, and 5 radiologists read these images under 2 different magnetic resonance imaging (MRI) sequences (Spin-echo and Cine MRI) to obtain the pre-experiment data. Then Obuchowski-Rockette method was used to estimate sample size. Results The mean area under curve (AUC) of aortic coarctation determined by radiologists was 0.941 (95% confidence interval ［CI］ 0.899-0.983) with the Spin-echo MRI sequence, and 0.897 (95% CI 0.837-0.957) with the Cine MRI sequence. When the effect size was 0.044 and the number of readers was 5, we needed 337 participants for random-reader random-case, 162 participants for fixed-reader random-case, and 282 participants for random-reader fixed-case. Conclusion In MRMC design, we need both the number of readers and cases; the larger the number of readers, the smaller the cases required. We need more samples under the situation of random-reader random-case, when the number of readers is ≥5.