It has been argued that much of human intelligence can be viewed as th
e process of matching stored patterns. In particular, it is believed t
hat chess masters use a pattern-based knowledge to analyze a position,
followed by a pattern-based controlled search to verify or correct th
e analysis. In this paper, a first-order system, called PAL, that can
learn patterns in the form of Horn clauses from simple example descrip
tions and general purpose knowledge is described. The learning model i
s based on (i) a constrained least general generalization algorithm to
structure the hypothesis space and guide the learning process, and (i
i) a pattern-based representation knowledge to constrain the construct
ion of hypothesis. It is shown how PAL can learn chess patterns which
are beyond the learning capabilities of current inductive systems. The
same pattern-based approach is used to learn qualitative models of si
mple dynamic systems and counterpoint rules for two-voice musical piec
es. Limitations of PAL in particular, and first-order systems in gener
al, are exposed in domains where a large number of background definiti
ons may be required for induction. Conclusions and future research dir
ections are given.