A criminal case will often involve various items of evidence of various kinds, related to each other in more or less complex ways. In such cases a suitable graphical diagram can be useful in understanding and analyzing evidence relationships. In recent years there has been much research by the statistics and artificial-intelligence communities on computerized methods for probability calculations in graphically specified problems, going under the name "probabilistic expert systems" (PES) as described by Buckleton and Walsh. Such systems, if developed further to incorporate the special features of forensic inference, could revolutionize the practice of forensic science by making feasible a full and reasoned assessment of the overall impact of the evidence. A PES is typically represented graphically by drawing a "node" for each variable in the problem and arrows between certain pairs of nodes. This paper uses an imaginary case example to show how the method enables dependencies between various aspects of the evidence to be considered. 4 figures and 14 references