The induction of regular languages (or associated recognizers) from examples has attracted much attention from researchers. The most part of the known methods for regular grammatical inference only use positive examples. Recently, new symbolic and neural approaches have been proposed to induce nite state automata from both positive and negative data. In this paper we present a type of Moore machines, that we call Unbiased Finite State Automata (or UFSA), which allow a symmetric processing of positive and negative information. Two new pseudo-incremental methods for regular grammatical inference from positive and negative examples using UFSA are described, which are based on an algorithm due to Oncina and Garcia. These methods, which work in polynomial time and improve the average time complexity of the original algorithm, always obtain a deterministic UFSA that is consistent with the given examples. A new feature of the proposed methods is the capability of generalizing negative information, in the same way that is usually done for positive data. Some experiments have been carried out to show the behaviour of the methods.