Objective Adults with Diabetes Mellitus (DM) experience high levels of depression and anxiety that are not always effectively ameliorated by current therapeutic approaches. The Self‐Regulatory Executive Function (S‐REF) model, which underpins metacognitive therapy (MCT), posits that depression and anxiety become persistent when stored metacognitive beliefs guide an individual to respond to common thoughts and feelings in a certain way. We hypothesized that (i) metacognitive beliefs would predict depression and anxiety independently of participants' representations of their illness; and (ii) rumination would mediate independent prediction of depression and anxiety by metacognitive beliefs. Design A prospective mediation study. Methods Four hundred and forty‐one adults with DM (Types 1 and 2) completed a two time‐point survey. Metacognitive beliefs, illness representations and rumination were measured at baseline, and depression and anxiety measured at baseline and 6‐months later. Data were analysed using structural equation modelling. Baseline illness representations, depression and anxiety were used as control variables. Results A structural equation analysis showed potential mediation, by baseline rumination, of any effects of baseline metacognitive variables on 6‐month distress in Type 1 and 2 diabetes samples. Significant standardized coefficients for relationships between the metacognitive latent variable and rumination were .67 (Type 1) and .75 (Type 2) and between rumination and distress of .36 and .43, respectively. These effects were independent of direct and independent effects of illness representation variables. Conclusions Findings are consistent with metacognitive beliefs playing a key role in depression and anxiety by increasing the likelihood of rumination in adults with DM. MCT may be an effective intervention for this population, subsequent to further longitudinal testing of the S‐REF model.