Best practices and pitfalls in multivariate pattern analysis of event-related potentials: A systematic review of preprocessing and analytical configurations

Multivariate pattern analysis (MVPA, decoding) has been increasingly used in event-related potential (ERP) research. This growing use reflects several advantages of decoding approaches, such as their ability to exploit high-dimensional spatiotemporal ERP features, detect subtle and distributed neural differences between conditions, avoid subjective electrode selection, and provide time-resolved insights into cognitive processing. However, decoding performance is highly sensitive to preprocessing choices and analysis settings, many of which are directly inherited from conventional ERP practices despite not always being optimal for decoding analyses. Evidence indicates that preprocessing steps can differentially affect univariate ERP measures and decoding outcomes. Similarly, decoding-specific analysis parameters, including classifier selection, trial construction strategies, overfitting control methods, cross-validation schemes, and evaluation metrics, can substantially shape decoding performance and its interpretation. Importantly, commonly used metrics such as accuracy and AUC, while practical and widely implemented in popular toolboxes, may introduce bias under conditions such as class imbalance, which are prevalent in ERP paradigms. Here, we present a systematic review of ERP decoding studies published between January 2021 and December 2025, providing a quantitative overview of prevailing preprocessing pipelines and decoding configurations. We integrate these results with empirical and methodological evidence to discuss the mechanisms, benefits, and potential risks associated with common analytical choices, highlighting key divergences between decoding-oriented and conventional ERP methodologies. Finally, we identify recurrent methodological pitfalls and offer practical recommendations aimed at improving transparency, reproducibility, and methodological validity in future ERP decoding research.