This paper provides a comprehensive literature review of neural network applications for option pricing and hedging, covering over 150 papers from the early 1990s to the present. The authors systematically compare papers based on six key characteristics: input features (e.g., moneyness, volatility estimates), output variables (option prices, implied volatilities, hedging ratios), benchmark models (e.g., Black-Scholes, binomial trees), performance measures (e.g., MAE, MSE, R²), data partition methods (chronological vs. random), and underlying assets (e.g., S&P 500, DAX, individual stocks). The review highlights that while most studies report positive performance of ANNs, many suffer from methodological flaws such as improper data partitioning that violates time series structure, leading to data leakage and overoptimistic results. The authors offer three pieces of advice for practitioners: use stationary input features, benchmark appropriately, and respect chronological order when splitting data. The paper also discusses related work where ANNs are used for model-based pricing (e.g., learning from simulated data) and regularization techniques like early stopping and dropout. This review serves as a valuable resource for researchers and practitioners by summarizing the state of the art and identifying best practices in the field.