Background: Detection of the mild ketosis induced by severely energy-restricted diets may be a clinically useful way to monitor and promote dietary adherence. Mild ketosis is often assessed using urine dipsticks, but accuracy for this purpose has not been tested. Objective: To determine the accuracy of urine dipsticks to detect mild ketosis during adherence to a severely energy-restricted diet. Methods: Two hundred and sixty three (263) fasting urine and 263 fasting blood samples were taken from 50 women (mean [standard deviation, SD] age 58.0 [4.3] years and body mass index 34.3 [2.4] kg/m 2) before and at six time points during or for up to 10 weeks after 16 weeks of severe energy restriction, achieved with a total meal replacement diet. The amount of ketones (acetoacetate) in the urine was classified as '0 (Negative)', '+/− (Trace)', '+ (Weak)' or '++ (Medium)' by urine dipsticks (Ketostix, Bayer). The concentration of ketones (β-hydroxybutyrate) in the blood was measured with our reference method, a portable ketone monitor (FreeStyle Optium, Abbott). The diagnostic accuracy of the urine dipsticks was assessed from the percent of instances when a person was actually 'in ketosis' (as defined by a blood β-hydroxybutyrate concentration at or above three different thresholds) that were also identified by the urine dipsticks as being from a person in ketosis (the percent 'true positives' or sensitivity), as well as the percent of instances when a person was not in ketosis (as defined by the blood monitor result) was correctly identified as such with the urine dipstick (the percent 'true negatives' or specificity). Thresholds of ≥0.3mM, ≥0.5mM or ≥1.0mM were selected, because mean blood concentrations of β-hydroxybutyrate during ketogenic diets are approximately 0.5mM. Sensitivity and specificity were then used to generate receiver operating characteristic curves, with the area under these curves indicating the ability of the dipsticks to correctly identify people in ketosis (1 = perfect results, 0.5 = random results). Radhika V Seimon and Amanda Sainsbury contributed equally to this work.