Uncertain butterflies are fundamental and popular graphlet motifs within uncertain bipartite networks, serving as a crucial metric in structural analysis. Despite extensive research have studied butterflies sufficiently on deterministic networks, few of works explore uncertain butterflies. In this paper, we introduce the Most Probable Maximum Weighted Butterfly (MPMB), which holds the highest probability of becoming a maximum weighted butterfly on an uncertain bipartite network. Proved that searching MPMBs is NP-Hard, we then proposed two samplingbased methods, namely Ordering Sampling (OS), and Ordering- Listing Sampling (OLS). The OS method is suitable for singletrial sampling, while the OLS method is optimized for multiple trials, which first finds candidate butterflies in rough before searching MPMBs. Our experimental results indicate that our basic method (OS) performs 1000x faster than the baseline and the optimized method (OLS) achieves another 180x speedup.