Pooling in recurrent neural architectures improves sequence classification by enhancing gradient flow and reducing positional biases. Mean-pooling, max-pooling, attention, and a new method called max-attention were compared. Pooling helps with learning in the early training stages and reduces bias towards the beginning and end of input sequences. This leads to significant performance gains in scenarios with limited resources and when important words are in the middle of the input. Max-attention often outperforms other pooling techniques in text classification tasks.