Over the past eight decades the ice-infested Northern Sea Route (NSR) has been steadily developed although historically it has been of little interest to commercial shipping companies. However, the shorter routes, more open waters, and the increasing demand for shipping have caused shipping companies to reconsider shipping along these routes. The NSR is now part of an overall world transportation system. However, a key problem of shipping in Arctic environments is the extreme low temperatures encountered and sea ice during most of the year. The American Bureau of Shipping (ABS) has provided some guidance for shipping along the Arctic sea routes with winterization of the Arctic going vessel as a suggested requirement. The technology and level of winterization is determined by the design service temperature (DST) estimated by the lowest mean daily average temperature (MDAT) over at least a 20-year period for the intended geographical area and season of operation. Systems that are more susceptible to the lower temperatures have another temperature which ABS designates as the Minimum Anticipated Temperature (MAT). However, the suitability of that value with respect to a risk based assessment on a vessel and its systems is questionable. For a modern risk-based approach to winterization, knowledge of the magnitude and frequency of occurrence for a given duration are prerequisites. This paper presents a new and more rational approach to estimate the DST or MAT. The approach is based on the set of Temperature-Duration-Frequency (TDF) curves developed for a given climate station. For this study, the Arctic was divided into four regions. Annual extreme low temperatures of durations 1, 3, 6, 12, 24, 36, 48, 72, 96, 120, 144, and 168 hours were extracted from the last 20 years of historical record from three to four representative stations in each region. Magnitudes of low temperatures were then estimated from the data for return periods ranging from 2 to 100 years. These estimates are then used to construct the TDF curves that provide estimates of the magnitude of extreme low temperatures for a given return period, for various durations of practical interest. It is proposed that the 100-year return period 1-hour duration lowest extreme temperature be used as the estimate of the MAT. The proposed approach addresses the drawbacks of the current approach and provides a practical estimate of the DST or MAT for risk-based winterization decisions.