The main advantage of threshold models of tumor recurrence is that the
y closely parallel the real observation process in cancer post-treatme
nt surveillance. Such models are typically based on the assumption tha
t a recurrent tumor becomes detectable when its size attains some thre
shold value which may be treated as a random variable. This paper disc
usses various stochastic models that have been proposed for the analys
is of data on tumor latency subject to censoring effects. A new model
allowing for surviving clonogenic cells is presented.