Recent studies have demonstrated that while neural ranking models excel in retrieval performance, they often exacerbate gender biases. This research addresses these biases by targeting three key sources: 1) training data, 2) neural embeddings, and 3) training strategies. I investigate data sampling techniques to mitigate gender bias in training data, methods for disentangling gender-related information from embeddings, and bias-aware training strategies that modify the loss function to penalize biased predictions. By tackling these sources of bias, my work aims to create more equitable information retrieval systems that preserve effectiveness while reducing the amplification of gender stereotypes.