Linear regression assumes a linear relationship, is sensitive to outliers, and may not perform well if the assumptions (like homoscedasticity or normality) are violated.