Judgement bias tasks are promising tools to assess emotional valence in animals, however current designs are often time-consuming and lack aspects of validity. This study aimed to establish an improved design that addresses these issues and can be used across species. Horses, rats, and mice were trained on a spatial Go/No-go task where animals could initiate each trial. The location of an open goal-box, at either end of a row of five goal-boxes, signalled either reward (positive trial) or non-reward (negative trial). Animals first learned to approach the goal-box in positive trials (Go) and to re-initiate/not approach in negative trials (No-go). Animals were then tested for responses to ambiguous trials where goal-boxes at intermediate locations were opened. The Go:No-go response ratio was used as a measure of judgement bias. Most animals quickly learned the Go/No-go discrimination and performed trials at a high rate compared to previous studies. Subjects of all species reliably discriminated between reference cues and ambiguous cues, demonstrating a monotonic graded response across the different cue locations, with no evidence of learning about the outcome of ambiguous trials. This novel test protocol is an important step towards a practical task for comparative studies on judgement biases in animals.