Control of granulation processes is usually achieved by measuring the torque and/or power draw of the mixer where agglomeration takes place. It is current industrial practice to determine the endpoint of granulation at a preset value of the torque or of the total electric power absorbed by the motor. While simple and easy to implement, this kind of control does not take into account explicitly the evolution of sizes during granule formation and growth. A somewhat different type of granulation monitoring is to measure stresses in the agglomerating mass by introducing a probe into the system. This method, while more direct, does not give significantly different results from torque measurements. The method proposed in the present work is based on direct measurements from a sensing element placed in the powder. It uses, however, stress fluctuations as input, instead of average stresses. The input signal is subsequently analyzed by applying Fourier transformations to get a qualitative understanding, while a quantitative analysis is carried out using a statistical method specially developed for this purpose. The statistics are based on the comparison of a reference time series of stress fluctuations with an ongoing time series measured during granulation. In this way, continuous monitoring land control) can be achieved that takes granule growth dynamics into account. To demonstrate the feasibility of the method, granulation experiments were performed in a fluidized bed in which shear is induced by a rotating, rough inner cylinder. The experimental system (described in detail in Apicella et al. [AlChE J. 43 (1997) 1362.]), also known as a Fluidized Couette Device (FCD), consists of two concentric cylinders with particles in the annulus. The main advantage of the FCD is the constant shear distribution generated within the powder mass. Some model experiments were also performed using a fine powder into which large particles of known size were introduced that replaced actual granules. Comparison of stress fluctuation signals from model experiments and from actual granulations with small and large aggregates showed good reproducibility and a correlation between runs with similar size granules.