Background: Transbronchial lung cryobiopsy (TBLC) is a novel technique for sampling lung tissue for interstitial lung disease diagnosis. The aim of this study was to establish the diagnostic accuracy of TBLC compared with surgical lung biopsy (SLB), in the context of increasing use of TBLC in clinical practice as a less invasive biopsy technique. Methods: COLDICE was a prospective, multicentre, diagnostic accuracy study investigating diagnostic agreement between TBLC and SLB, across nine Australian tertiary hospitals. Patients with interstitial lung disease aged between 18 and 80 years were eligible for inclusion if they required histopathological evaluation to aid diagnosis, after detailed baseline evaluation. After screening at a centralised multidisciplinary discussion (MDD), patients with interstitial lung disease referred for lung biopsy underwent sequential TBLC and SLB under one anaesthetic. Each tissue sample was assigned a number between 1 and 130, allocated in a computer-generated random sequence. Encoded biopsy samples were then analysed by masked pathologists. At subsequent MDD, de-identified cases were discussed twice with either TBLC or SLB along with clinical and radiological data, in random non-consecutive order. Co-primary endpoints were agreement of histopathological features in TBLC and SLB for patterns of definite or probable usual interstitial pneumonia, indeterminate for usual interstitial pneumonia, and alternative diagnosis; and for agreement of consensus clinical diagnosis using TBLC and SLB at MDD. Concordance and κ values were calculated for each primary endpoint. This study is registered with the Australian New Zealand Clinical Trials Registry, ACTRN12615000718549. Findings: Between March 15, 2016, and April 15, 2019, we enrolled 65 patients (31 [48%] men, 34 [52%] women; mean age 66·1 years [SD 9·3]; forced vital capacity 83·7% [SD 14·2]; diffusing capacity for carbon monoxide 63·4% [SD 12·8]). TBLC (7·1 mm, SD 1·9) and SLB (46·5 mm, 14·9) samples were each taken from two separate ipsilateral lobes. Histopathological agreement between TBLC and SLB was 70·8% (weighted κ 0·70, 95% CI 0·55–0·86); diagnostic agreement at MDD was 76·9% (κ 0·62, 0·47–0·78). For TBLC with high or definite diagnostic confidence at MDD (39 [60%] of 65 cases), 37 (95%) were concordant with SLB diagnoses. In the 26 (40%) of 65 cases with low-confidence or unclassifiable TBLC diagnoses, SLB reclassified six (23%) to alternative high-confidence or definite MDD diagnoses. Mild-moderate airway bleeding occurred in 14 (22%) patients due to TBLC. The 90-day mortality was 2% (one of 65 patients), following acute exacerbation of idiopathic pulmonary fibrosis. Interpretation: High levels of agreement between TBLC and SLB for both histopathological interpretation and MDD diagnoses were shown. The TBLC MDD diagnoses made with high confidence were particularly reliable, showing excellent concordance with SLB MDD diagnoses. These data support the clinical utility of TBLC in interstitial lung disease diagnostic algorithms. Further studies investigating the safety profile of TBLC are needed. Funding: University of Sydney, Hunter Medical Research Institute, Erbe Elektromedizin, Medtronic, Cook Medical, Rymed, Karl-Storz, Zeiss, and Olympus.