Normality Assumption in ANOVA
The normality assumption says that the ANOVA errors are normally distributed:
$$ \varepsilon_{ij} \sim \mathcal{N}(0, \sigma^2). $$In practice this is checked with a normal QQ plot of residuals. Strong systematic departures from the line indicate that the assumption may be questionable.