The morphological properties of genealogical and marriage alliance networks constitute a key to the understanding of matrimonial behavior and social norms, in particular where these norms have not been explicitly formalized. Their analysis, however, faces a major difficulty: the actual datasets which allow researchers to reconstruct kinship and alliance networks are generally subject to a marked observer bias, if only due to limitations of observer mobility and/or informant memory. This paper presents an agent based simulation method destined to evaluate the impact of this bias on some key indicators of kinship and alliance networks (such as matrimonial circuit frequencies). The method consists in explicitly simulating the exploration of a given network by a virtual observer, the bias being introduced by the observer's inclination for choosing informants who are more or less closely related to each other. The article presents the model for genealogical and for alliance networks, applies it to a series of artificial networks exhibiting some characteristic morphological patterns, and discusses the divergence of observed from real patterns for different kinds and degrees of observer bias. The methods presented have been implemented in the free software Puck 2.0.