AIM--To question the observer reliability or agreement of reports on the intranodal and extranodal tumour growth patterns in early metastasised non-small cell lung cancer (NSCLC). METHODS--In a pilot study original histological sections of mediastinal lymph node metastases from NSCLC obtained by lymph node dissection (n = 82) or by mediastinoscopy (n = 62) were examined and classified independently by three pathologists as extranodal, intranodal, or indefinite. After clear criteria for these growth patterns had been defined sections were re-examined and recategorised one year later. Interobserver agreement was examined for both investigations. RESULTS--In the dissected lymph nodes the kappa value improved significantly from 0.52 (moderate agreement) at the first investigation to 0.72 (good agreement) at the second. In the mediastinoscopic lymph node biopsy specimens an increase in kappa value from 0.50 at the first to 0.67 at the second examination was found, although this improvement was not significant. In mediastinoscopic biopsy specimens a very high proportion of tissue samples showed indefinite tumour extension. CONCLUSION--Good reproducibility of intranodal and extranodal growth patterns in the histological examination of mediastinal lymph node metastases can be achieved, provided that pathologists use strictly defined criteria. In mediastinoscopic biopsy specimens it is often impossible to differentiate between intranodal and extranodal tumour growth.