Diagnosis of SARS-CoV-2 infections is mostly based on the nasopharyngeal swabs (NPS). However, this collection is invasive and uncomfortable, especially for children and patients with coagulopathies, whose NPS collection often causes bleeding. Thus, the aim of this study was to evaluate the usefulness and accuracy of saliva for the diagnosis of COVID-19 in patients presenting bleeding disorders. Samples of NPS, oropharyngeal swabs (OPS), and saliva were collected simultaneously from 1159 hospitalized patients with hematological diseases and from 524 healthcare workers, both symptomatic and asymptomatic for SARS-CoV-2. All samples were evaluated for SARS-CoV-2 by qRT-PCR. SARS-CoV-2 was detected in NPS, OPS and saliva from 16.9%, 14.4% and 15.6% individuals, respectively. Tests in saliva showed sensitivity, specificity, and overall agreement of 73.3%, 96.9% and 92.7% (=0.74), respectively. Salivary tests had good accuracy (AUC = 0.7) for discriminating negative and positive qRT-PCR for SARS-CoV-2. Higher sensitivity was observed in symptomatic than in non-symptomatic patients, as well as in healthy subjects than in patients with hematological disease, in both OPS and saliva. The mean viral load in NPS was significantly higher than in OPS and in saliva samples (p < 0.001). Saliva is a good diagnostic tool to detect SARS-CoV-2, especially among patients symptomatic for COVID-19, and is a valuable specimen for mass screening of hospitalized patients with hematological diseases, especially for those that with bleeding disorders.