Abstract Background Lipoprotein (a) (Lp(a)) measurement may help guide CVD risk prediction, is thought to be causal in several CVD outcomes, and phase 3 intervention trials of Lp(a) lowering agents are underway. We aimed to investigate the population attributable fraction due to elevated Lp(a) and its utility in CVD risk prediction. Methods In 413,724 participants from UK Biobank, associations of serum Lp(a) with composite fatal/nonfatal CVD (n=10,065 events), fatal CVD (n=3247), coronary heart disease (n=16,649), ischaemic stroke (n=3191), and peripheral vascular disease (n=2716) were compared using Cox models. Predictive utility was determined by C-index changes. The population attributable fraction was estimated. Results Median Lp(a) was 19.7nmol/L (interquartile interval 7.6–75.3nmol/L). 20.8% had Lp(a) values >100nmol/L; 9.2% had values >175nmol/L. After adjustment for classical risk factors, in participants with no baseline CVD and not taking a statin, 1 standard deviation increment in log Lp(a) was associated with a HR for fatal/nonfatal CVD of 1.09 (95% CI 1.07–1.11). Associations were similar for fatal CVD, coronary heart disease, and peripheral vascular disease. Adding Lp(a) to a prediction model containing traditional CVD risk factors improved the C-index by +0.0017 (95% CI 0.0009, 0.0026). We estimated that having Lp(a) values >100nmol/L accounts for 5.7% of CVD events in the whole cohort. We modelled that an ongoing trial to lower Lp(a) in patients with CVD and Lp(a) above ∼175nmol/L may be expected to reduce CVD risk by 20.3%, assuming causality, and an achieved Lp(a) reduction of 80%. Conclusions Population screening for elevated Lp(a) may help to predict CVD and target Lp(a) lowering drugs to those with markedly elevated levels, if such drugs prove efficacious. Population attributable fractions: Lp(a) Funding Acknowledgement Type of funding source: Other. Main funding source(s): Chest, Heart, and Stroke Association Scotland and British Heart Foundation