A need exists to identify dehydrated individuals under stressful settings beyond the laboratory. A predictive index based on changes in saliva osmolality has been proposed, and its efficacy and sensitivity was appraised across mass (water) losses from 1 to 7%. Twelve euhydrated males [serum osmolality: 286.1 mOsm kg -1 H 2O (SD 4.3)] completed three exercise- and heat-induced dehydration trials (35.6°C, 56% relative humidity): 7% dehydration (6.15 h), 3% dehydration (with 60% fluid replacement: 2.37 h), repeat 7% dehydration (5.27 h). Expectorated saliva osmolality, measured at baseline and at each 1% mass change, was used to predict instantaneous hydration state relative to mass losses of 3 and 6%. Saliva osmolality increased linearly with dehydration, although its basal osmolality and its rate of change varied among and within subjects across trials. Receiver operating characteristic curves indicated a good predictive power for saliva osmolality when used with two, single-threshold cutoffs to differentiate between hydrated and dehydrated individuals (area under curve: 3% cutoff = 0.868, 6% cutoff = 0.831). However, when analysed using a double-threshold detection technique (3 and 6%), as might be used in a field-based monitor, <50% of the osmolality data could correctly identify individuals who exceeded 3% dehydration. Indeed, within the 3-6% dehydration range, its sensitivity was 64%, while beyond 6% dehydration, this fell to 42%. Therefore, while expectorated saliva osmolality tracked mass losses within individuals, its large intra- and inter-individual variability limited its predictive power and sensitivity, rendering its utility questionable within a universal dehydration monitor. © 2012 Springer-Verlag.