SUMMARY1. An attempt has been made to census elephant by a method which will enable the reliability of the results to be assessed.2. The method used was a system of stratified random samples.3. Stratification was based on the positions of elephant clumps recorded on a reconnaissance flight.4. Strip samples were selected at random for the various strata and elephant and other large mammals were counted in a strip of 440 m width from an aircraft travelling 100 m above the ground at 130 km/h.5. The problems of maintaining a fixed height above the ground and counting animals in a strip of constant width are acknowledged.6. One stratum was totally counted for purposes of comparison.7. Elephant, eland, kongoni, Grant's gazelle, buffalo, oryx, zebra, impala, ostrich and giraffe have been counted.8. The results are set out in a series of tables and estimates and confidence intervals have been computed.9. A discussion is held on the influences of various factors on the width of sample strips observed from the aircraft.10. The total count and sample count of stratum 1 are observed to be in fairly close agreement.11. The re‐stratification of stratum 1 to take advantage of the observed clumps of elephant reduced the standard error of the estimate proportionally. The estimate was increased because the samples were no longer valid for the new strata.12. The elephant population is of the order of 9,000 animals, as was predicted from observation flights, and the confidence interval of ±1,974 is acceptable for a first attempt at a sample count for elephant. The Mkomazi elephant were observed in three distinguishable groups, this being a distribution pattern frequently recorded since 1965.13. Estimates and confidence intervals for other mammals do not refer to the whole region. It is observed that standard errors as a proportion of estimates are proportional to the degree of clumping shown by the species.14. The census took 45 h, and cost K£800. It was accomplished in 7 d by a single pilot and observer. The Mkomaz