Health risks, defined in terms of the probability that an individual w
ill suffer a particular type of adverse health event within a given ti
me period, can be understood as referencing either natural entities or
complex patterns of belief which incorporate the observer's values an
d knowledge, the position adopted in the present paper. The subjectivi
ty inherent in judgements about adversity and time Frames can be easil
y recognised, but social scientists have tended to accept uncritically
the objectivity of probability. Most commonly in health risk analysis
, the term probability refers to rates established by induction, and s
o requires the definition of a numerator and denominator. Depending up
on their specification, many probabilities may be reasonably postulate
d for the same event, and individuals may change their risks by decidi
ng to seek or avoid information. These apparent absurdities can be und
erstood if probability is conceptualised as the projection of expectat
ion onto the external world. Probabilities based on induction from obs
erved frequencies provide glimpses of the future at the price of accep
tance of the simplifying heuristic that statistics derived from aggreg
ate groups can be validly attributed to individuals within them. The p
aper illustrates four implications of this conceptualisation of probab
ility with qualitative data from a variety of sources, particularly a
study of genetic counselling for pregnant women in a U.K. hospital. Fi
rstly, the official selection of a specific probability heuristic refl
ects organisational constraints and values as well as predictive optim
isation. Secondly, professionals and service users must work to mainta
in the facticity of an established heuristic in the face of alternativ
es. Thirdly, individuals, both lay and professional, manage probabilis
tic information in ways which support their strategic objectives. Four
thly, predictively sub-optimum schema, for example the idea of AIDS as
a gay plague, may be selected because they match prevailing Social va
lue systems. (C) 1998 Elsevier Science Ltd. All rights reserved.