Non-ideal bundle models that predict the effect of filament slack, misalignment and bonding were compared to the predicted ideal bundle stress-strain response and the measured response of an Alumina tow (Nextel(TM) 610) as a function of exposure temperature. The ideal bundle model assumes that the filaments are perfectly aligned, independent, and filament strength is characterized by single filament tests; the non-ideal bundle models relax these assumptions. The Nextel(TM) 610 tow was found to behave ideally until the sintering bonds between filaments were strong enough to resist fracture during testing. When bonded filament clusters existed throughout testing, the weakest filament in the cluster likely caused failure of the entire cluster. Therefore, the assumption of independence was not valid, and the measured bundle strength was lower than the ideal bundle prediction. The assumption that the strength distribution of the bonded filaments was the same as the filaments removed from bonded filament clusters also appears to be invalid. It is hypothesized that the effective decrease in strength of bonded filaments is a result of induced shear stresses that form along the bond line when bonded filaments are either bent or twisted. However, within a composite this motion is limited. The modeling indicated that 10% clustering can be tolerated with little degradation in composite properties as long as the clusters remain independent from their neighbors. (C) 2003 Kluwer Academic Publishers.