Target distribution refers to the probability distribution that a statistical model aims to approximate or learn from during training. It is crucial in supervised learning, where the model's performance is evaluated based on how well it can predict or match this distribution from the given data inputs.