Positron emission tomography (PET) images are potential quantitative biomarkers. Understanding long-term (months/years) biomarker variability is important for establishing confidence intervals on studies using such biomarkers over these time frames. PET biomarkers are derived from activity concentration () extracted from PET images. Over 30 months, we measured the stability of decay-normalized counts () and by scanning the same 4.5–cm-diameter Ge-68 cylinder weekly, the same Na-22 point source daily, and a refilled 20-cm F-18 cylinder phantom monthly on a clinical TOF-PET/CT scanner. Longitudinal and adjacent-measurement variability was characterized. We found no drift in or for properly calibrated images over 24 months. During this time, ranged to 6% for count-matched Ge-68 and F-18 images, with coefficient of variation (COV) across time of 2.3% (Ge-68, 81 scans) and 3.2% (F-18, 24 scans). At typical patient image count levels the Ge-68 () COV across time was 6.9% (9.6%). Changes in between adjacent F-18 scans () ranged between , with corresponding date-matched changes in Ge-68 ranging . We recommend (1) tracking trends in with image as a check of quantitative data corrections/calibrations and (2) tracking both mean and COV of (single time point measures) to hundredths precision using standardized uptake values.