Prospective case-control studies investigating lipoprotein(a) [Lp(a)] as a risk factor for atherosclerosis have measured Lp(a) in samples stored frozen up to nearly 20 years. We therefore prospectively examined the influence of long-term plasma sample storage on measured values, depending on the molecular weight of apolipoprotein(a) [apo(a)] isoforms. Apo(a) phenotyping was performed in 310 plasma samples, and Lp(a) was measured after 3 and 28 months of storage at -80 degrees C. The values of both measurements correlated significantly for both low- and high-molecular-weight apo(a) phenotypes (r = .97 and r = .98, respectively, P < .001). Nevertheless, we detected on average a small decrease of 4.83% from mean +/- SD (median) 21.24 +/- 23.54 (11.10) mg/dL to 20.02 +/- 21.72 (10.55) mg/dL, which was statistically significant (P < .001). The absolute and relative Lp(a) decrease over time became larger with a decreasing number of kringle IV repeats of apo(a) (P < .05), and Lp(a) decreased markedly more in subjects with low-molecular-weight compared with those with high-molecular-weight apo(a) isoforms (-3.26 versus -0.46 mg/dL, P < .05). More than 70% of the absolute Lp(a) decrease in the total sample was caused by samples with low-molecular-weight apo(a) isoforms, which represented only 27% of the sample. Low-molecular-weight apo(a) isoforms are reportedly more frequent in patients with atherothrombotic disease compared with control subjects. Measurement of Lp(a) in several-year-old frozen samples is therefore likely to result in a preferential decrease and false lower Lp(a) concentrations in patient groups compared with control groups. The negative results of some prospective studies with retrospective measurement of Lp(a) may be caused by such an artifact.