A recent model <Kok and Tauxe, 1996a> that attempts to explain asymmetrical sawtooth patterns of relative paleointensity from sediments <Valet and Meynadier, 1993> as the consequence of cumulative long-term viscosity effects has been tested theoretically and experimentally. A mathematical description of the model demonstrates that only a sharp distribution of relaxation times would be able to match the paleomagnetic record, which implies that all the magnetic grains would be identical. This has direct consequences for the distribution of the blocking temperatures (Tb), which would not exceed 260 ¿C, and for the coercivities of the natural remanent magnetization (NRM). This Tb spectrum derived from the model is inconsistent with the unblocking temperatures of the NRM and with the Tb spectrum of the experimental thermoremanent magnetization (TRM) of the sediment that recorded the sawtooth. New results including thermal demagnetization and Thellier-Thellier experiments across Brunhes-Matuyama confirm the overall stability of the original records. The stability of the older part of the record (2 to 4 Ma) has also been strengthed. The sawtooth is unchanged between 20 mT and 65 mT, and thermal demagnetization in a neutral atmosphere at 300 ¿C and 450 ¿C duplicates the results obtained by alternation field (a.f.) demagnetization. Thus viscous processes, at least within the framework of a cumulative viscous model and for the existing data sets, are not responsible for saw-toothed paleointensity records. ¿ 1998 American Geophysical Union