Partial‐mouth scores are often used in epidemiologie studies to estimate the prevalence and severity of periodontal disease. Such estimates systematically underestimate the prevalence of disease, but the bias can be in either direction for disease severity. Three large data sets containing full‐mouth examinations (mesiobuccal, midbuccal, distobuccal, and midlingual (MBDL) sites per tooth) for either attachment loss or probing pocket depths were used to investigate the magnitudes of systemic error that occur by employing four specific partialmouth scores (M, MB, MBD, and MBDL sites per tooth), which are based on the random half‐mouth technique. For prevalence of disease the sensitivity of a partial score was a function of the disease level in the population. All four partial scores were sensitive enough to adequately portray true prevalence using the 2‐mm demarcation value, the MBD and MBDL scores might still be considered adequate for the 4‐mm value, but none of these partial scores were adequate for the 7‐mm value. For disease severity the MBDL score produced unbiased estimates, the others were biased. The relative biases for the MB and MBD scores were all under 10% (in absolute value), but the M score produced severe relative biases, 24% for probing pocket depths and ‐12% for attachment loss.