Urothelial carcinoma (UC) is the most common type of bladder cancer. Its treatment depends from both tumour extension (stage) and aggressiveness (grade). The gold standard for detecting UC is white-light cystoscopy, followed by tissue biopsy and pathological examination for determining tumour stage and grade. However, such process is invasive, time-consuming and prone to sampling errors. In this framework, optical spectroscopy techniques provide fast, label-free and non-invasive alternatives to standard histopathology. Thus, we combined auto-fluorescence, diffuse reflectance and Raman spectroscopies in a compact and transportable setup based on an optical fibre-probe. The latter was coupled to three laser diodes (emitting at 378 nm, 445 nm and 785 nm) and to a halogen lamp for exciting and collecting autofluorescence, Raman and reflectance spectra, respectively. This experimental setup was used for studying fresh biopsies of urothelial tumour (82 samples) and healthy bladder (32 samples) collected from 49 patients undergoing Transurethral Resection of Bladder Tumours (TURBT). All spectral recordings were done within 30 minutes from surgical resection, and optical inspection required less than 2 minutes for each sample. The recorded data were analysed using Principal Component Analysis (PCA) for obtaining an automated classification of the examined samples based on the intrinsic spectral information provided by all three techniques. We found that multimodal spectroscopy provides high-sensitivity, high-specificity discriminating capability for UC detection, grading and staging. The presented strategy generates results similar to gold standard histology, but in a fast and label-free way, offering the potential for endoscopic in vivo applications.