Multimodal Sentiment Analysis (MSA) has emerged as a pivotal domain within Artificial Intelligence (AI), focusing on comprehending sentiments by leveraging multiple data modalities. Despite the advancements, existing deep neural network approaches primarily concentrate on augmenting feature extraction techniques and refining the architectures of sentiment analysis frameworks. Such emphasis often overlooks the critical aspect of how various modalities interact and mutually influence the analysis results. Addressing this gap, we introduce a novel approach by integrating Bayesian Network Fusion (BNF) for multimodal sentiment analysis. Our model, distinguished by its Bayesian Network (BN) inspired fusion mechanism, strategically combines audio, visual, and textual data through distinct neural network pathways. This decision-level fusion process not only elevates the model’s interpretability but also significantly enhances its predictive accuracy. We rigorously evaluate our approach on three multimodal sentiment analysis datasets, where our model demonstrates superior classification performance and offers profound insights into the intricate interplay among different modalities.