We examined the diagnostic accuracy of human touch (HT) method in assessing hypothermia against axillary digital thermometry (ADT) by a trained non-medical field investigator (who supervised activities of community health volunteers) in seven villages of Agra district, Uttar Pradesh, India. Body temperature of 148 newborns born between March and August 2005 was measured at four points in time for each enrolled newborn (within 48 h and on days 7, 30 and 60) by the field investigator under the axilla using a digital thermometer and by HT method using standard methodology. Total observations were 533. Hypothermia assessed by HT was in agreement with that assessed by ADT (<36.5 degrees C) in 498 observations. Hypothermia assessed by HT showed a high diagnostic accuracy when compared against ADT (kappa 0.65-0.81; sensitivity 74%; specificity 96.7%; positive predictive value 22; negative predictive value 0.26). HT is a simple, quick, inexpensive and programmatically important method. However, being a subjective assessment, its reliability depends on the investigator being adequately trained and competent in making consistently accurate assessments. There is also a need to assess whether with training and supervision even the less literate mothers, traditional birth attendants and community health volunteers can accurately assess mild and moderate hypothermia before promoting HT for early identification of neonatal risk in community-based programs.