In the course of salmonella epizootics in guinea-pigs due to Salm. enteritidis and Salm. typhi-murium, respectively, salmonella conjunctivitis was widespread, and conjunctival swabbing was found to be a more effective procedure for detecting the spread of infection than the examination of rectal swabs. Eye-swabbing with segregation of eye-positive guinea-pigs and their cage-mates was apparently successful in controlling two Salm. enteritidis epizootics.

Later experimental observations showed that the conjunctival route was a far more effective one for producing systemic salmonella infection in guinea-pigs than the oral route. The 50% spleen infecting dose by the conjunctival route was of the order of 100 organisms, while a dose of 100,000,000 organisms of the same strain of Salm. enteritidis was ineffective by mouth. Blood-borne invasion from the primary conjunctival lesion occurred by the lymphatic route. Gut infection of guinea-pigs infected experimentally throughout the eye was secondary to spleen infection and probably occurred by excretion of salmonellae through the gall-bladder.

The importance of the conjunctival route in natural salmonella infections of guinea-pigs was confirmed by goggling experiments, which showed that, except in animals with complicating conditions such as pregnancy, effectively goggled guinea-pigs were highly resistant to infection in environments heavily contaminated with Salm. enteritidis which produced high infection rates in ungoggled control animals.

I am very grateful to my technician, Mr G. Sandys, for his enthusiastic help, particularly with the goggling experiments.

1. The relationship has been examined between the variance and the mean of pock counts made by a number of different workers.

2. In all cases the variance was considerably greater than that expected if the distributions were of the Poisson form. Recent work of Westwood et al. (1957) suggests that variation approaching the Poisson expectation can be achieved by appropriate technical measures.

3. With one exception, there appeared to be a linear relationship between the logarithm of the variance and the logarithm of the mean. The slope of this line varied from one laboratory to another, but not, apparently, from one occasion to another in any one laboratory.

4. The hypothesis that for any set of egg membranes the expectations of the pock counts bear constant ratios to each other, irrespective of the virus density, was untenable: the observed variances increase less rapidly with high counts than would have been expected on this theory.

5. Statistical methods are proposed for comparing and combining sets of counts at different dilutions. If Poisson variability were achieved, simpler methods would be available.

I am much indebted to Dr L. Collier, Prof. A. W. Downie, Dr C. Kaplan and Dr F. O. McCallum for permission to use their unpublished data; to Dr J. C. N. Westwood for discussing with me his (then) unpublished work; to Miss Irene Allen for computing assistance; and to Mrs G. M. Young for the preparation of the diagrams.

1. The membrane piece technique for in vitro titrations of the infectivity of influenza virus is described. Rectangles of shell, about 8 × 25 mm., with the chorio-allantoic membrane still attached (membrane pieces) are cut from thirteenth-day fertile eggs. One piece in a test-tube with glucose-buffered salt solution forms an individual assay unit. Five or more tubes are inoculated with each virus dilution. After incubation at 37° C. for 72 hr., with agitation for the first 24 hr. the fluid in each tube is tested for haemagglutinins. From the results at each dilution, an estimate of the 50% membrane piece (MP50) infectivity titre is obtained.

2. Six hundred assay units, with pieces cut from twenty eggs, can be set up by two workers in 1 hr. and used for titration of between three and twenty-four individual virus preparations, depending on the reliability desired for the 50% end-point estimates.

3. With the D.S.P. and PR 8 strains of influenza A virus, the MP50 titres parallel the EID50 titres from egg titrations, but are eight times and twenty times lower, respectively. The MP50: EID50 ratio is the same for various preparations of the same strain, including standard allantoic fluid and chorio-allantoic membrane virus, incomplete virus, and inactivated (heated) allantoic fluid virus. Preliminary experiments with Lee influenza B virus show that slightly different experimental conditions are required, and the MP50 titres are about fifty times less than the EID50 titres.

4. Consistent results have been obtained on titration of samples of the same virus preparation on a number of occasions over a period of several months.

5. A large number of membrane pieces can be used to test each virus dilution, and sampling variations in the MP50 estimates thus made quite small. Statistical data on the reliability of a 50 % titration result, and on the minimum significant differences between two end-points, are given for different values of n, the number of membrane pieces used to test each virus dilution, and of d, the log dilution step.

We are grateful to Mr J. Collins for invaluable technical assistance, and also to Miss I. Allen for help with the computations.