Accuracy of titration may be taken as a measure of the reliability of the methods and work of a laboratory.With such a small number of titrations as four, a convincing mean, representing the most probable true titre, is hard to calculate. A fairly close approximation, however, is possible in the form of the geometrical mean.It is also possible to correct for constant local factors which cause the titrations of a given laboratory to run above or below the mean. By all methods of titration a surprisingly large proportion of large errors are recorded.The most consistent results, with the smallest proportion of large errors, are obtained by the prearranged standard method with a single standardized suspension. The H titres are rather more consistent than the O; probably because H agglutination is easier than O to read correctly.Local methods give greater differences than the agreed (standard) method. Since the H suspensions used in one laboratory were also sensitive to O agglutination the supposed H titres are unreal, and cannot be treated statistically.Local O methods give a considerably greater range of error than the standard method, as is shown by a comparison of the mean errors of Tables II and IV, and by the percentages of errors over and under 100 per cent given in Table VI.The full titres (H or O or both) of the serums measured by local routine Widal methods, with living suspensions in two laboratories and killed ones in the other two, show a considerably greater proportion of large differences than are shown by the titrations with the standard method. They do not, however, differ more widely than the titres obtained with local O suspensions.The results of the laboratories that use killed H and O suspensions in routine work are more consistent and accurate than those of the laboratories that use living suspensions.It is suggested that gross technical mistakes are a commoner cause of error than is usually supposed. Errors of dilution of the serums and other technical variations must also be responsible for a part of the very wide range of differences observed.The use of old strains, tending to roughness, is to be avoided.