Solving systems of polynomial equations in an ultimate way means to nd the isolated primes of the associated variety and to present them in a way that is well suited for further computations.[5] proposes an algorithm, that uses several Grobner basis computations for a dimension reduction argument, delaying factorization to the end of the algorithm. In this paper we i n v estigate the opposite approach, heavily using factorization (of multivariate polynomials), delaying the computation of stable ideal quotients. At a heuristic level this is exactly the well known Grobner algorithm with factorization and constraint inequalities, available in all major computer algebra systems. In a preceding paper [9] we reported on some experience with a new version of this algorithm, implemented in our REDUCE package CALI [8]. Here we discuss, how this approach may bere ned to produce triangular systems in the sense of [12] and [13]. Such a re nement guarantees, di erent to the usual Grobner factorizer, to produce a quasi prime decomposition, i.e. the resulting components are at least pure dimensional radical ideals. As in [9] our method weakens the usual restriction to lexicographic term orders.