A learning-curve property, originally prescribed for describing reliab
ility growth when time to failure is observed, is applied to discrete
reliability-growth processes for systems for which only discrete succe
ss or failure events are observed (eg, a missile or torpedo). Derivati
ons leading to a well-known discrete reliability-growth model are pres
ented, and the role of implicit assumptions relating to testing strate
gy is identified. An alternative testing strategy, particularly approp
riate for destructive test of very expensive systems, is proposed and
a new discrete reliability-growth model is derived. Extensions to both
types of discrete reliability-growth models, which account for the di
stinction between assignable cause (readily correctable failure modes)
and non-assignable cause (state-of-the-art failure modes), are also p
rovided. Parameter estimates for each of the models can be obtained vi
a usual maximum likelihood procedures. Each of the derived discrete re
liability-growth models is legitimate in the sense that ft accommodate
s monotonic growth in reliability as testing progresses through consec
utive system configurations. The use of any particular discrete reliab
ility-growth model cannot be supported by the learning-curve property,
however, unless the appropriate assumptions identified in this paper
are satisfied.