The range of sample sizes within which linear chromatographic behavior is achieved in a column depends on the surface heterogeneity of the RPLC adsorbents. Two widely different commercial adsorbents were tested, the end-capped XTerra-C18 and the non-end-capped Resolve-C18. Adsorption isotherm data of caffeine were acquired by frontal analysis. These data were modeled and used to calculate the adsorption energy distribution (AED). This double analysis informs on the degree of surface heterogeneity. The best adsorption isotherm models are the bi-Langmuir and the tetra-Langmuir isotherms for XTerra and Resolve, respectively. Their respective AEDs are bimodal and quadrimodal distributions. This interpretation of the results and the actual presence of a low density of high-energy adsorption sites on Resolve-C18 were validated by measuring the dependence of the peak retention times on the size of caffeine samples (20-microL volume, concentrations 10, 1, 0.1, 1 x 10(-2), 1 x 10(-3), 1 x 10(-4), and 1 x 10(-5) g/L). The experimental chromatograms agree closely with the band profiles calculated from the best isotherms. On Resolve-C18, the retention time decreases by 40% when the sample concentration is increased from 1 x 10(-5) to 10 g/L. The decrease is only 10% for Xterra-C18 under the same conditions. The upper limit for linear behavior is 1 x 10(-4) g/L for the former adsorbent and 0.01 g/L for the latter. The presence of a few high-energy adsorption sites on Resolve-C18, with an adsorption energy 20 kJ/mol larger than that of the low-energy sites while the same difference on Xterra is only 5 kJ/mol, explains this difference. The existence of adsorption sites with a very high energy for certain compounds affects the reproducibility of their retention times and a rapid loss of efficiency in a sample size range within which linear behavior is incorrectly anticipated.