The objective of the study was to determine the agreement between rectal, axillary and inguinal temperatures and to estimate the accuracy of these measurements in detecting hyperthermia and hypothermia in dogs presented at a veterinary teaching hospital in the tropical Guinea Savannah zone. Prospectively, body temperature was measured in 610 dogs, using digital thermometry in the axillary, inguinal and rectal regions. Overall, axillary and inguinal temperatures significantly underestimated rectal temperature, with a mean difference of -0.39 ± 0.02°C (95% confidence interval: -0.43 to -0.35; limit of agreement: -1.27 to 0.49) and - 0.34 ± 0.02°C (95% confidence interval, -0.37 to -0.30; limit of agreement: -1.15 to 0.47), respectively. The limits of agreement of axillary and inguinal temperatures were wide and above the pre-determined maximal acceptable difference of ±0.50°C recommended for clinical significance of rectal temperature in dogs. Bland-Altman plots showed that the confidence intervals of the mean differences of axillary and inguinal temperatures did not include the value zero, thereby indicating that the tested methods lack agreement with rectal temperature. Sensitivity and specificity for the detection of hyperthermia with axillary temperature were 72.1% and 30.5%, respectively. In contrast, sensitivity and specificity for the detection of hyperthermia with inguinal temperature were 77.9% and 26.2%, respectively. The magnitude of disagreement between axillary, inguinal and rectal temperatures was affected by age, breed and sex being slightly lower in mature, non-native breed and female dogs. Axillary and inguinal temperature measurements in dogs significantly underestimated rectal temperature measurements by -0.39 ± 0.02°C and -0.34 ± 0.02°C, respectively. The results indicate that axillary and inguinal temperatures should not be used as a replacement for rectal temperature due to the wide limits of agreement. In addition, axillary and inguinal temperatures may not be suitable in detecting hyperthermia because the sensitivity were lower than the required set-point of 90.0% for clinical identification of hyperthermia.