Recurrent neural networks (RNNs) are designed to model temporal dependencies in sequential data, such as video frames, by constructing directed graphs that unfold over time. This chapter explores RNN architectures, including bidirectional RNNs, which incorporate past and future context, and deep RNNs, which stack multiple layers for enhanced modeling. Long short-term memory (LSTM) units, with their gating mechanisms, address the vanishing gradient problem and improve long-term dependency learning. Bidirectional LSTMs further extend this by combining forward and backward sequences. The chapter also introduces gated recurrent units (GRUs), a simplified LSTM variant, balancing efficiency and performance. Finally, the application of RNNs in video classification is highlighted through the CNN + LSTM framework, which combines spatial feature extraction with temporal modeling for accurate video analysis. This chapter provides a concise overview of RNNs and their role in sequential data modeling.