Abstract： Objective:To verify the accuracy of a non-invasive hemoglobinometer based on photoelectric plethysmography (PPG) .Methods:Partial least squares (PLS) algorithm for machine learning and PPG technique were used to develop a non-invasive hemoglobinometer. Thereafter, the hemoglobin (Hb) levels in a cohort of voluntary blood donors and patients in Department of Hematology of PLA General Hospital were measured with the newly developed non-invasive device and recorded as spectrophotometric hemoglobin (SpHb) . For these subjects, Hb levels were also measured by invasive testing and recorded as total Hb (tHb) for reference. The data were processed using the MATLAB R2015b software. The correlation coefficient ( R) , root mean square error ( RMSE) , residual, Clarke grid error analysis, and the Bland-Altman (BA) analysis by MedCalc19.5.6 were used to verify the accuracy of the hemoglobinometer. Results:Referring to the commonly used invasive detection in clinical setting, the correlation coefficient between the two testing was 0.915 2, and the RMSE was 0.959 51. The results of Clarke error grid analysis showed all data were situated in zones A (88.82%) and B (11.18%) , with none in zone C. BA agreement test showed that the mean difference between SpHb and tHb was 0.7 g/L, 95% confidence interval ( CI) -18.1-19.5 g/L. Conclusion:The non-invasive hemoglobinometry based on PPG can rapidly and accurately measure the hemoglobin level of patients, and is promising as an alternative to the conventional invasive hemoglobin testing.