As the great majority of gene expression (GE) biodosimetry studies have been performed using blood as the preferred source of tissue, searching for simple and less-invasive sampling methods is important when considering biodosimetry approaches. Knowing that whole saliva contains an ultrafiltrate of blood and white blood cells, it is expected that the findings in blood can also be found in saliva. This human in vivo study aims to examine radiation-induced GE changes in saliva for biodosimetry purposes and to predict radiation-induced disease, which is yet poorly characterized. Furthermore, we examined whether transcriptional biomarkers in blood can also be found equivalently in saliva. Saliva and blood samples were collected in parallel from radiotherapy (RT) treated patients who suffered from head and neck cancer (n = 8) undergoing fractioned partial-body irradiations (1.8 Gy/fraction and 50–70 Gy total dose). Samples were taken 12–24 h before first irradiation and ideally 24 and 48 h, as well as 5 weeks after radiotherapy onset. Due to the low quality and quantity of isolated RNA samples from one patient, they had to be excluded from further analysis, leaving a total of 24 saliva and 24 blood samples from 7 patients eligible for analysis. Using qRT-PCR, 18S rRNA and 16S rRNA (the ratio being a surrogate for the relative human RNA/bacterial burden), four housekeeping genes and nine mRNAs previously identified as radiation responsive in blood-based studies were detected. Significant GE associations with absorbed dose were found for five genes and after the 2nd radiotherapy fraction, shown by, e.g., the increase of CDKN1A (2.0 fold, P = 0.017) and FDXR (1.9 fold increased, P = 0.002). After the 25th radiotherapy fraction, however, all four genes (FDXR, DDB2, POU2AF1, WNT3) predicting ARS (acute radiation syndrome) severity, as well as further genes (including CCNG1 [median-fold change (FC) = 0.3, P = 0.013], and GADD45A (median-FC = 0.3, P = 0.031)) appeared significantly downregulated (FC = 0.3, P = 0.01–0.03). A significant association of CCNG1, POU2AF1, HPRT1, and WNT3 (P = 0.006-0.04) with acute or late radiotoxicity could be shown before the onset of these clinical outcomes. In an established set of four genes predicting acute health effects in blood, the response in saliva samples was similar to the expected up- (FDXR, DDB2) or downregulation (POU2AF1, WNT3) in blood for up to 71% of the measurements. Comparing GE responses (PHPT1, CCNG1, CDKN1A, GADD45A, SESN1) in saliva and blood samples, there was a significant linear association between saliva and blood response of CDKN1A (R2 = 0.60, P = 0.0004). However, the GE pattern of other genes differed between saliva and blood. In summary, the current human in vivo study, (I) reveals significant radiation-induced GE associations of five transcriptional biomarkers in salivary samples, (II) suggests genes predicting diverse clinical outcomes such as acute and late radiotoxicity as well as ARS severity, and (III) supports the view that blood-based GE response can be reflected in saliva samples, indicating that saliva is a “mirror of the body” for certain but not all genes and, thus, studies for each gene of interest in blood are required for saliva.