An order-recursive algorithm is presented for the generalised eigendec
omposition of definite Hermitian pencils. An important feature is that
it consists of a number (equal to the current order) of independent i
terative eigenvalue searches followed by direct computation of the cor
responding eigenvectors. The independence aspect produces the potentia
l for parallel hardware implementations. The order-recursive nature of
the algorithm results in the potential for solving problems of an a p
riori unknown, adequate minimal size rather than a single maximum size
only. Such algorithms may be useful for sensor-array signal processin
g, for example in early direction-of-arrival estimation; The order-rec
ursive nature is not detrimental to numerical performance, as it is sh
own that the algorithm is competitive with standard algorithms solving
the single maximum-size problem only.