Saliva sampling could serve as an alternative non-invasive sample for SARS-CoV-2 diagnosis while rapid antigen tests (RATs) might help to mitigate the shortage of reagents sporadically encountered with RT-PCR. Thus, in the RESTART study we compared antigen and RT-PCR testing methods on nasopharyngeal (NP) swabs and salivary samples. We conducted a prospective observational study among COVID-19 hospitalized patients between 10 December 2020 and 1 February 2021. Paired saliva and NP samples were investigated by RT-PCR (Cobas 6800, Roche-Switzerland, Basel, Switzerland) and by two rapid antigen tests: One Step Immunoassay Exdia ^® COVID-19 Ag (Precision Biosensor, Daejeon, Korea) and Standard Q ^® COVID-19 Rapid Antigen Test (Roche-Switzerland). A total of 58 paired NP-saliva specimens were collected. A total of 32 of 58 (55%) patients were hospitalized in the intensive care unit, and the median duration of symptoms was 11 days (IQR 5-19). NP and salivary RT-PCR exhibited sensitivity of 98% and 69% respectively, whereas the specificity of these RT-PCRs assays was 100%. The NP RATs exhibited much lower diagnostic performance, with sensitivities of 35% and 41% for the Standard Q ^® and Exdia ^® assays, respectively, when a wet-swab approach was used (i.e., when the swab was diluted in the viral transport medium (VTM) before testing). The sensitivity of the dry-swab approach was slightly better (47%). These antigen tests exhibited very low sensitivity (4% and 8%) when applied to salivary swabs. Nasopharyngeal RT-PCR is the most accurate test for COVID-19 diagnosis in hospitalized patients. RT-PCR on salivary samples may be used when nasopharyngeal swabs are contraindicated. RATs are not appropriate for hospitalized patients.