Glucose concentrations were determined in capillary and venous blood and in venous plasma from 100 consecutive patients referred to an oral glucose tolerance test. The capillary blood was immediately transferred to a haemolysing and glucose stabilizing solution and frozen within 4h. Venous blood was drawn in heparin-sodium fluoride tubes and stored for 0, 4 or 24h at 4 or 20 degrees C. Aliquots of the venous blood were then treated in the same way as the capillary blood. The primary tubes were centrifuged, and aliquots of the plasma were stabilized, as described. All specimens from the same sampling event were analysed in the same analytical series on EBIO compact. Deming linear regression equations y = a + bx were calculated to estimate the glucose concentration in one specimen from that in a differently treated specimen e.g.: B(cB:fPt)---Glucose (0h) = 0.61 + 0.897*P(vB;fPt)-Glucose (24h. 20 C). n = 100, SD(y/x) = 0.25 mmol/L and B(vB) Glucose (0h) = 0.53 + 0.897*P(vB)--Glucose (24h. 20 degrees C), n = 196, SD(y/x) = 0.26 mmol,/L. The non-analytical part of the standard deviation of the ratio between the near-patient test result and its predicted value was 0.038 for both of the above predictions. In conclusion, the analytical (including sampling) variation of near-patient tests of glucose can be assessed by laboratory analysis on mailed, heparin-fluoride stabilized venous blood or on plasma samples made from the mailed blood.