According to the current dogma, cholesteryl ester transfer protein (CETP) decreases high-density lipoprotein (HDL)-cholesterol (C) and increases low-density lipoprotein (LDL)-C. However, detailed insight into the effects of CETP on lipoprotein subclasses is lacking. Therefore, we used a Mendelian randomization approach based on a genetic score for serum CETP concentration (rs247616, rs12720922 and rs1968905) to estimate causal effects per unit (µg/mL) increase in CETP on 159 standardized metabolic biomarkers, primarily lipoprotein subclasses. Metabolic biomarkers were measured by nuclear magnetic resonance (NMR) in 5672 participants of the Netherlands Epidemiology of Obesity (NEO) study. Higher CETP concentrations were associated with less large HDL (largest effect XL-HDL-C, P = 6 × 10^–22) and more small VLDL components (largest effect S-VLDL cholesteryl esters, P = 6 × 10^–6). No causal effects were observed with LDL subclasses. All these effects were replicated in an independent cohort from European ancestry (MAGNETIC NMR GWAS; n ~20,000). Additionally, we assessed observational associations between ELISA-measured CETP concentration and metabolic measures. In contrast to results from Mendelian randomization, observationally, CETP concentration predominantly associated with more VLDL, IDL and LDL components. Our results show that CETP is an important causal determinant of HDL and VLDL concentration and composition, which may imply that the CETP inhibitor anacetrapib decreased cardiovascular disease risk through specific reduction of small VLDL rather than LDL. The contrast between genetic and observational associations might be explained by a high capacity of VLDL, IDL and LDL subclasses to carry CETP, thereby concealing causal effects on HDL.