The pollination syndrome hypothesis predicts that plants pollinated by the same pollinator group bear convergent combinations of specific floral functional traits. Nevertheless, some studies have shown that these combinations predict pollinators with relatively low accuracy. This discrepancy may be caused by changes in the importance of specific floral traits for different pollinator groups and under different environmental conditions. To explore this, we studied pollination systems and floral traits along an elevational gradient on Mount Cameroon during wet and dry seasons. Using Random Forest (Machine Learning) models, allowing the ranking of traits by their relative importance, we demonstrated that some floral traits are more important than others for pollinators. However, the distribution and importance of traits vary under different environmental conditions. Our results imply the need to improve our trait-based understanding of plant-pollinator interactions to better inform the debate surrounding the pollination syndrome hypothesis.