Fuzzy relational identi"cation builds a relational model describing a system's behaviour by a nonlinear mapping between its variables. In this paper, we propose a new fuzzy relational algorithm based on the simpli"ed max}min relational equation. The algorithm presents an adaptation method applied to the gravity-centre of each fuzzy set based on the error integral value between the measured and predicted system's output, and uses the concept of time-variant universe of discourse. The identi"cation algorithm also includes a method to attenuate the noise in#uence in the extracted system's relational model using a fuzzy "ltering mechanism. The algorithm is applied to a one-step forward prediction of a simulated and experimental motor drive system. The identi"ed model has its input}output variables (stator-reference current and motor speed signal) treated as fuzzy sets, whereas the relations existing between them are described by means of a matrix R de"ning the relational model extracted by the algorithm. The results show the good potentialities of the algorithm in predicting the behaviour of the system and in attenuating through the fuzzy "ltering method possible noise distortions in the relational model.