Suicide risk is high in early psychosis. Recent research has suggested that mood variability may be associated with levels of suicidal thoughts and behaviour. This has not been investigated in individuals during and following a first or second episode of non-affective psychosis. Repeated-measures data over 18 months from a large randomised controlled trial for cognitive behaviour therapy (N=309) were analysed using latent growth curve modelling, whereby both the variability and the level of depression, anxiety and guilt were entered as predictors of suicidality. The variability of depression, but not guilt and anxiety, predicted the course of suicidality even when controlling for a large range of potential confounders. The level of depression, anxiety and guilt for each participant also strongly predicted the development of suicidality. The findings support the theory that variability in depression may contribute to the formation of suicidal ideation and related behaviour. More variable depression may be harder to predict and intervene against, and therefore increase the likelihood that suicidality escalates. The levels of emotions may also be an important determinant. This has implications for the treatment and assessment of suicidality in early psychosis.