Outlier detection based on granular-ball (GB) is used to efficiently identify abnormal samples in complex data, especially when dealing with fuzzy and uncertain data. However, the existing outlier detection methods are mainly based on the coarse-grained input of GBs. Although these methods are robust and efficient, they often suffer from significant information loss. To address this issue, we propose a fuzzy rough outlier detection method in two-layer GB space. Firstly, we transform the universe into a structure, consisting of a coarse-grained GB layer and a fine-grained object layer, using GB computing and fuzzy rough sets. Secondly, by combining the advantages of both fine-grained and coarse-grained approaches, we map a single object to the GB layer to calculate fuzzy counting. On this basis, we construct an attribute sequence and calculate the outlier metric under each attribute space. Correspondingly, we design an algorithm to implement this approach. Finally, through experiments on open datasets, we compare our algorithm with seven other algorithms. The results demonstrate that our proposed method achieves good performance and strong robustness.