Nuclear magnetic resonance T1-T2 maps are popular for characterizing fluids in shale. The characterization, however, is often done manually, which is difficult for shale due to its complicated nature. This work investigates the clustering approach for fluid characterization on T1-T2 maps by comparing 6 different algorithms: k-means, Gaussian mixture model, spectral clustering, and 3 hierarchical methods. T1-T2 maps are collected on shale samples at as-received and dried conditions. We propose two cluster validity indices to select the optimal cluster number and best algorithm. Our results validate the capability of the two indices. Gaussian mixture model is found to be the best algorithm in most of the cases, as its fluid partitioning shows the highest consistency with theoretical fluid boundaries. In addition, 5 fluid components are identified from Gaussian mixture model, and their values are qualitatively validated by comparing with those in literature. Results also indicate that clustering is sensitive to the fluid distribution. Drying the sample producing better clustering by revealing the footprint of organic matter. This work provides a practical guide for applying cluster analysis in fluid characterization in Nuclear magnetic resonance T1-T2 maps.