Linear regression assumes a linear relationship between independent and dependent variables, independence of observations, constant variance of errors (homoscedasticity), normally distributed residuals, and no high correlation between independent variables (no multicollinearity).