The results of clinical biochemistry tests are used in the diagnosis of diabetes mellitus by comparing them with reference values or agreed international criteria. Low analytical imprecision is required so that reference values are not unduly widened by analytical variability. Most laboratory test results are used in monitoring. Low analytical imprecision is required so that changes seen in sequential results reflect stability, amelioration or deterioration, not simply analytical variability. Analytical bias should be absent so that results are (1) comparable in individuals over time as methods evolve and instrumentation changes, (2) equivalent over locale since tests may be done in practitioners' surgeries, outpatient clinics, and laboratories, and (3) able to be interpreted against the fixed criteria. A plethora of strategies have been proposed to set numerical desirable standards of performance, these being termed analytical goals, including use of reference values, opinions of clinicians, state of the art, the views of experts, and data on biological variation. Use of data on within-subject biological variation is currently considered the best approach. A generally applicable goal is that the total error should be less than one-half of the within-subject biological variation. Achievement of this adds about 10% through analytical variability to the true test variability. Analytical goals (CV, %) for glucose, fructosamine, urinary albumin, and haemoglobin A1 assays are 2.2, 2.1, 18, and 3.3%, respectively. These goals are targets worthy of attainment, not inflexible criteria of acceptance or rejection of methods.