Roller compaction is a dry granulation process used to convert powder blends into free flowing agglomerates. During scale up or transfer of roller compaction process, it is critical to maintain comparable ribbon densities at each scale in order to achieve similar tensile strengths and subsequently similar particle size distribution of milled material. Similar ribbon densities can be reached by maintaining analogous normal stress applied by the rolls on ribbon for a given gap between rolls. Johanson (1965) developed a model to predict normal stress based on material properties and roll diameter. However, the practical application of Johanson model to estimate normal stress on the ribbon is limited due to its requirement of accurate estimate of nip pressure i.e. pressure at the nip angle. Another weakness of Johanson model is the assumption of a fixed angle of wall friction that leads to use of a fixed nip angle in the model. To overcome the above mentioned limitations, we developed a novel approach using roll force equations based on a modified Johanson model in which the requirement of pressure value at nip angle was eliminated. An instrumented roll on WP120 roller compactor was used to collect normal stress data measured at three locations across the width of a roll (P1, P2, P3), as well as gap and nip angle data on ribbon for placebo and various active blends along with corresponding process parameters. The nip angles were estimated directly using experimental pressure profile data of each run. The roll force equation of Johanson model was validated using normal stress, gap, and nip angle data of the placebo runs. The calculated roll force values compared well with those determined from the roll force equation provided for the Alexanderwerk(®) WP120 roller compactor. Subsequently, the calculation was reversed to estimate normal stress and corresponding ribbon densities as a function of gap and RFU (roll force per unit roll width). A placebo model was developed and calibrated using a subset of placebo run data obtained on WP120. The roll force values were calculated using vendor supplied equation. The nip angle was expressed as a function of gap and RFU. The nip angle, gap and RFU were used in a new roll force equation to estimate normal stress P2 at the center of the ribbon. Using ratios P1/P2 and P3/P2 from the calibration data set, P1 and P2 were estimated. The ribbon width over which P1, P2, and P3 are effective was determined by minimizing sum square error between the model predicted vs. experimental ribbon densities of the calibration set. The model predicted ribbon densities of the placebo runs compared well with the experimental data. The placebo model also predicted with reasonable accuracy the ribbon densities of active A, B, and C blends prepared at various combinations of process parameters. The placebo model was then used to calculate scale up parameters from WP120 to WP200 roller compactor. While WP120 has a single screw speed, WP200 is equipped with a twin feed screw system. A limited number of roller compaction runs on WP200 was used as a calibration set to determine normal stress profile across ribbon width. The nip angle equation derived from instrumented roll data collected on WP120 was applied to estimate nip angles on WP200 at various processing conditions. The roll force values calculated from vendor supplied equation and the nip angle values were used in roll force equation to estimate normal stress P2 at the tip of the feed screws. Based on feed screw design, it was assumed that the normal stress at the center of the ribbon was equal to those calculated at the tip of the feed screws. The ratio of normal stress at the edge of the ribbon Pe to the normal stress P2 at the feed screw tip was optimized to minimize sum square error between model predicted vs. experimental ribbon densities of the calibration set. The model predicted ribbon densities of the batches prepared on WP200 compared well with the experimental data thus indicating success of the scale up procedure. For the demonstration purpose, the model was also calibrated using instrumented roll data of active C batches. This would be applicable when sufficient amount of API is available or placebo model cannot predict ribbon density of active batches.