Philosophers disagree how abstract, theoretical principles can be appl
ied to instances. This paper generates a puzzle for law theorists, cau
sal theorists and inductivists alike. Intractability can force scienti
sts to use a ''semi-empirical'' method, in which some of an equation's
theoretically-determinable parameters are replaced with values taken
directly from the data. This is not a purely deductive or inductive pr
ocess, nor does it involve causes and capacities in any simple way (Hu
mphreys 1995). I argue the predictive successes of such methods requir
e us to reanalyze our views about the nature of prediction, the status
of models, and the goal(s) of science. When laws and experimental evi
dence are neither individually nor jointly sufficient for prediction,
models become the locus of understanding. I analyze an historically im
portant debate about the use of semi-empirical methods to construct po
tential energy surfaces.