Nasopharyngeal swabs are considered the preferential collection method for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) diagnostics. Less invasive and simpler alternative sampling procedures, such as saliva collection, are desirable. We compared saliva specimens and nasopharyngeal (NP) swabs with respect to sensitivity in detecting SARS-CoV-2. A nasopharyngeal and two saliva specimens (collected by spitting or oral swabbing) were obtained from >2500 individuals. All samples were tested by RT-qPCR, detecting RNA of SARS-CoV-2. The test sensitivity was compared on the two saliva collections with the nasopharyngeal specimen for all subjects and stratified by symptom status and viral load. Of the 2850 patients for whom all three samples were available, 105 were positive on NP swab, whereas 32 and 23 were also positive on saliva spitting and saliva swabbing samples, respectively. The sensitivity of the RT-qPCR to detect SARS-CoV-2 among NP-positive patients was 30.5% (95% CI, 1.9%-40.2%) for saliva spitting and 21.9% (95% CI, 14.4%-31.0%) for saliva swabbing. However, when focusing on subjects with medium to high viral load, sensitivity on saliva increased substantially: 93.9% (95% CI, 79.8%-99.3%) and 76.9% (95% CI, 56.4%-91.0%) for spitting and swabbing, respectively, regardless of symptomatic status. Our results suggest that saliva cannot readily replace nasopharyngeal sampling for SARS-CoV-2 diagnostics but may enable identification of the most contagious cases with medium to high viral loads.