Random glucose is widely used in routine clinical practice. We investigated whether this non-standardized glycemic measure is useful for individual diabetes prediction.The Study of Health in Pomerania (SHIP), a population-based cohort study in north-east Germany, included 3107 diabetes-free persons aged 31–81 years at baseline in 1997–2001. 2475 persons participated at 5-year follow-up and gave self-reports of incident diabetes. For the total sample and for subjects aged ≥50 years, statistical properties of prediction models with and without random glucose were compared.A basic model (including age, sex, diabetes of parents, hypertension and waist circumference) and a comprehensive model (additionally including various lifestyle variables and blood parameters, but not HbA1c) performed statistically significantly better after adding random glucose (e.g., the area under the receiver-operating curve (AROC) increased from 0.824 to 0.856 after adding random glucose to the comprehensive model in the total sample). Likewise, adding random glucose to prediction models which included HbA1c led to significant improvements of predictive ability (e.g., for subjects ≥50 years, AROC increased from 0.824 to 0.849 after adding random glucose to the comprehensive model+HbA1c).Random glucose is useful for individual diabetes prediction, and improves prediction models including HbA1c.