The aim of this study is to establish and test an in vitro digestion-in situ absorption model that can mimic in vivo drug flux by employing a physiologically relevant value of the membrane surface area (S)/volume (V) ratio for accurate prediction of oral drug absorption from lipid-based formulations (LBFs). Three different types of LBFs (Type IIIA-MC, Type IIIA-LC, and Type IV) loaded with cinnarizine (CNZ), a lipophilic weak base with borderline permeability, and a control suspension were prepared. Subsequently, a simultaneous in vitro digestion-permeation experiment was conducted using a side-by-side diffusion cell with a dialysis membrane having a low S/V value. During digestion, CNZ partially precipitated for Type IV, while it remained solubilized in the aqueous phase for Type IIIA-MC and Type IIIA-LC in the donor compartment. However, in vitro drug fluxes for Type IIIA-MC and Type IIIA-LC were lower than those for Type IV due to the reduced free fraction of CNZ in the donor compartment. In pharmacokinetic studies, a similar improvement in in vivo oral exposure relative to suspension was observed, regardless of the LBFs used. Consequently, a poor correlation was found between in vitro permeation and areas under the plasma concentration-time curve (AUCoral) (R2 = 0.087). A luminal concentration measurement study revealed that this discrepancy was attributed to the extremely high absorption rate of CNZ in the gastrointestinal tract compared to that across a dialysis membrane evaluated by the in vitro digestion-permeation model, i.e., the absorption of CNZ in vivo was completed regardless of the extent of the free fraction, owing to the rapid removal of CNZ from the intestine. Subsequently, we aimed to predict the oral absorption of CNZ from the same formulations using a model that demonstrated high drug flux by employing the physiologically relevant S/V value and rat jejunum segment as an absorption sink (for replicating in vivo intestinal permeability). Predigested formulations were injected into the rat intestinal loop, and AUCloop values were calculated from the plasma concentration-time profiles. A better correlation was found between AUCloop and AUCoral (R2 = 0.72), although AUCloop underestimated AUCoral for Type IV due to the precipitation of CNZ during the predigestion process. However, this result indicated the importance of mimicking the in vivo drug absorption rate in the predictive model. The method presented herein is valuable for the development of LBFs.