The accurate monitoring of climate change imposes strict requirements upon observing systems, in particular regarding measurement accuracy and long-term stability. Currently available data records of the essential climate variables (temperature-T, geopotential-p, humidity-RH, wind, and cloud properties) in the upper-air generally fail to fulfill such requirements. This raises serious issues about the ability to detect, quantify and understand recent climate changes and their causes. GCOS is currently implementing a Reference Upper-Air Network (GRUAN) in order to fill this major void within the global observing system. As part of the GRUAN implementation plan we provide herein fundamental guidelines for establishing and maintaining reference quality atmospheric observations which are based on principal concepts of metrology, in particular traceability. It is argued that the detailed analysis of the uncertainty budget of a measurement technique is the critical step for achieving this goal. As we will demonstrate with an example, detailed knowledge of the calibration procedures and data processing algorithms are required for determining the uncertainty of each individual data point. Of particular importance is the careful assessment of the uncertainties introduced by correction schemes adjusting for systematic effects.