Objectives: To estimate the long-term true change variation ('signal') and short-term within-person variation ('noise') of the different lipid measures and evaluate the best measure and the optimal interval for lipid rescreening. Design: Retrospective cohort study from 2005 to 2008. Setting: A medical health check-up programme at a centre for preventive medicine in a teaching hospital in Tokyo, Japan. Participants: 15 810 apparently healthy Japanese adults not taking cholesterol-lowering drugs at baseline, with a mean body mass index of 22.5 kg/m2 (SD 3.2). Main outcome measures: Annual measurement of the serum total cholesterol (TC), low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and calculation of the ratio of TC/HDL and LDL/HDL. Measurement of the ratio of long-term true change variation ('signal') to the short-term within-person variation ('noise') for each measure. Results: At baseline, participants (53% male) with a mean age of 49 years (range 21-92) and a mean TC level of 5.3 mmol/l (SD 0.9 mmol/l) had annual check-ups over 4 years. Short-term within-person variations of TC, LDL, HDL, TC/HDL, and LDL/HDL were 0.12 (coefficient of variation (CV) 6.4%), 0.08 (CV 9.4%), 0.02 (CV 8.0%) mmol2/l2, 0.08 (CV 7.9%) and 0.05 (CV 10.6%), respectively. The ratio of signal-to-noise at 3 years was largest for TC/HDL (1.6), followed by LDL/HDL (1.5), LDL (0.99), TC (0.8) and HDL (0.7), suggesting that cholesterol ratios are more sensitive re-screening measures. Conclusion: The signal-to-noise ratios of standard single lipid measures (TC, LDL and HDL) are weak over 3 years and decisions based on these measures are potentially misleading. The ratios, TC/HDL and LDL/HDL, seem to be better measures for monitoring assessments. The lipid rescreening interval should be >3 years for those not taking cholesterol-lowering drugs.