In relational databases, normalization is the process of organizing data to reduce redundancy and improve data integrity. However, this can lead to multiple tables being joined in queries, which can degrade performance. Denormalization involves merging tables or adding redundant columns to enhance query performance at the cost of increased redundancy and potential inconsistencies. It is often used in read-heavy applications, data warehousing, and systems that require high query performance.