There is a substantial body of evidence showing that substance use is associated with mental health problems, but establishing whether associations are causal and whether specific substances have independent effects is more challenging. The studies in this thesis aimed to investigate whether cannabis and cigarette use are associated with psychotic experiences, depression and anxiety, and examine the extent to which associations might be due to confounding, bias and reverse causation. Data used were from ALSPAC, a large longitudinal birth cohort. For most analyses exposure measures were assessed when participants were aged 16 years, and outcome measures at 18 years. Regression models were used to assess associations before and after adjustment for a large variety of potential confounders. Multiple imputation was used to try and account for potential attrition bias. A Mendelian randomisation approach was used to assess whether associations between smoking and psychotic experiences are likely to be causal, using SNP rsl051730 as an instrumental variable. Cannabis and cigarette use were associated with all mental health outcomes in unadjusted analyses. Adjustment for pre-birth and childhood confounding made little difference to most analyses. Adjustment for other substance use led to substantial attenuation in many of the analyses investigated. Although there appeared to be some divergence between the effects of different substances on specific outcomes, and differences in the specific effects of substances between outcomes, this was not supported by statistical evidence. There was no evidence to support reverse causation. The Mendelian randomisation analysis did not provide evidence of a causal association between smoking and psychotic experiences, but was underpowered. The findings are consistent with previous literature from longitudinal studies that show associations between both cannabis and cigarette use with incident mental health outcomes, but also indicate that unmeasured confounding likely affects most studies to date. At present there is not enough evidence to be confident that associations between substance use and mental health are causal. Future research should utilise novel methods to approach these questions from multiple angles in order to progress the field.