Genetic birth defects often show seemingly random variability of occurrence or of severity between individuals. These are termed incomplete penetrance and variable expressivity, respectively. Variability occurs even within monozygotic twins, where most sources of genetic variation are absent. This poses problems clinically and has been difficult to explain. Archetypal examples of unpredictable variability occur in the development of the enteric nervous system (ENS). In Hirschsprung disease the distal ENS fails to form, and in slow-transit constipation the distal ENS is present but is dysfunctional. We have developed mathematical cellular automata models of ENS formation and unpredictable but distally-amplified variabilities resembling (Hirscshprung) or allowing (slow-transit constipation) these conditions emerged spontaneously. These results are in silico examples of incomplete penetrance and variable expressivity. These differences in outcome at the level of the entire system between replications with identical initial conditions and identical rules of behaviour for ENS agents (cells) were driven by stochastic choice of local motile and proliferative behaviour of each of the many agents, iterated over many of cycles. Surprisingly, given the number of agents and iterations, the entire system did not ‘balance out’ to end identically. These examples show in detail how the inability to totally determine specific aspects of cells’ morphogenetic behaviour can result in incomplete penetrance and variable expressivity in birth defects. Transposed into biological systems, these mechanisms therefore impose limits on disease predictability.