The asymptotic variances of various estimators of the population size
from a recapture study are calculated. The estimation methods compared
are maximum likelihood and a class of natural martingale estimators w
ith both time variation and behavioural response. It is found that (i)
the martingale methods appropriate for time variation or behavioural
response are fully efficient, (ii) there is no loss of efficiency at a
ll in allowing for time variation, (iii) there is large loss of effici
ency in allowing for behavioural response, (iv) there is a natural mar
tingale method that allows for both time variation and behavioural res
ponse.