The Year 2000 edition of the Highway Capacity Manual (HCM) will contain several new chapters providing guidance to planners on how the HCM can be used in transportation planning models. This paper describes a new speed-flow curve recommended by the new HCM for use in planning models and illustrates the application of this new speed-flow curve in the San Francisco Bay Area. This new speed-flow curve is called the "Akcelik" curve. Previous research by Dowling, Singh and Cheng demonstrated that the Akcelik speed-flow model produces significantly more accurate speed estimates than the standard Bureau of Public Roads (BPR) equation traditionally used in planning models. As has been shown in the wealth of speed-flow data gathered for freeways and other facilities, the Akcelik speed-flow curve is relatively insensitive to increases in traffic volumes until volumes approach capacity. Then the speeds predicted by the Akcelik curve drop fairly rapidly (at the rate predicted by queuing theory). The Akcelik predicted speeds, however, do not go as low as those predicted by the standard BPR curve for extreme volume/capacity ratios (greater than 2.00 v/c). This is because the Akcelik curve has the property of maintaining a linear increase in link travel times for v/c ratios greater than 1.00. This linear increase in travel times for v/c ratios greater than 1.00 is also in accordance with queuing theory and has been born out in simulation model results. This paper shows a specific application of the Akcelik curve to the San Francisco Bay Area. It was found that the Akcelik curve did not adversely affect equilibrium assignment model run times. The computation times were similar for both the Akcelik and a variation of the BPR currently used by the Metropolitan Transportation Commission. The Akcelik curve, however, resulted in a significant (and the authors believe more realistic) lowering of the estimated mean systemwide speed by 3 to 6 mph (5 to 10 km/h). The Akcelik curve also results in more realistic assigned traffic volumes that tend to cluster more closely around a v/c ratio of 1.00 with much fewer links with v/c ratios in excess of 1.5 than obtained with traditional BPR curves and their variants.