Knowledge of flowering and leaf flush phenology is important for a deep understanding of the responses of ecosystem functions, ecosystem services, and biodiversity to climate change. Process-based phenology models, which account statistically for chilling accumulation for endodormancy release and thermal accumulation after endodormancy release, can predict flowering and leaf flush dates, but they are difficult to develop for poorly studied species. A simpler approach is needed. We propose a model of the blooming phenology of Yoshino cherry trees (Cerasus ×yedoensis) based on bidirectional self-organizing maps (SOM). SOM, a data mining approach, is a tool for categorizing patterns in n-dimensional observation data by forming a two-dimensional lattice. The bi-directional SOM was applied to the data of daily mean temperature, using flowering (or full blooming) dates as teacher signals. We inputted daily mean air temperatures during the thermal accumulation period (mainly from February to just before flowering) into multiple input vectors and obtained flowering or full bloom dates into an output vector. The mean absolute errors between predicted and observed dates at 42 locations in Japan ranged from 2.8 to about 4.5 days for flowering and full bloom. The bidirectional SOM approach had a slightly higher error than the process-based phenology model approach, but it has an advantage in predicting flowering and leaf flush dates of poorly studied species under future warming conditions.