The quick delivery of a functionally truncated product is one of the most common results in iterative development, and has become the predominant development approach. One of its drawbacks is the appearance of incomplete artifacts between iterations. Consequently, well-known size-estimation methods can not be used in iterative development. This paper addresses the problem of size estimation in iterative development. We present a novel approach that enables early size estimation using Unified Modeling Language (UML) artifacts. The approach incorporates self-improvement steps that increase the estimation accuracy in subsequent iterations. The demonstration of its applicability and research results are also presented. The results anticipate the possibility of a significant improvement in size and effort estimates by applying the approach presented here.