In situ model experiments of milk bubbles have been successfully used to simulate the previously proposed mechanism by which pores tend to become situated at two grain junctions in the final stage of sintering. Ever since Brook [1] developed a model of interactions between a spherical pore and a single boundary, a lot of work has been done, not only to conduct a more rigorous analysis of the breakaway problem, but also to work out the microstructural evolution based on this model [2-5]. However, as far as the microstructural evolution is concerned, it has been recognized that pore-boundary separation is not a primary cause of the initiation of abnormal grain growth. Detailed discussions have been reported recently by Fang [6]. Moreover, three things to be emphasized are: first, at the very beginning of the final stage of sintering, most pores in the uniform microstructures are located on the three-grain junctions rather than on the two-grain junctions, shown in Fig. 1 [7]; secondly, according to the theoretical calculation [4, 5], pores on the twograin junctions have little effect on the boundary velocity, because separation typically precedes the onset of drag [4, 8] and pores at three-grain junctions cannot become isolated within grains (in the absence of major instabilities) the pores must first be displaced from three-grain junctions onto two-grain interfaces [4, 8]; and thirdly, two-sided pores are not randomly distributed and the mechanisms for developing this kind of pore have long been neglected. Regarding how pores located in the three-grain junctions become located in the two-grain junctions, Fang and Palour [9] have proposed two mechanisms to describe this transition: the first opportunity occurs when the cylindrical pores break down through ovulation [10, 1 1]; and the second opportunity occurs when one of the surrounding grains becomes relatively large, as shown in Fig. 2. Moreover, the larger the difference of size between this larger grain and the other two is, the greater the possibility that this mechanism would occur. Although the latter mechanism is indirectly supported by some final microstructures [9], it is inadequate to describe the whole process. In this study a delicate model experiment was carried out to provide insight into the latter mechanism. The milk bubbles were produced by pouring milk from a bottle of chocolate milk with a small mouth. It should be noted that the milk ought to be cold enough to make the patterns of bubbles change slowly. The advantages of the use of milk bubbles are that: first, although the bubbles sticking to the wall of a bottle are twodimensional, actually their behaviour is a type of three-dimensional one, which is the reason why we can see the two-sided bubbles shown in Fig. 3; secondly, the boundaries are clearer than these of soap bubbles; and thirdly, the time taken for the patterns of those bubbles to change is not too slow. A sequence of changing patterns of bubbles is shown in Figs 3 and 4. In the first figure the bubble indicated by the white arrow is considered as a simulated pore. This configuration is quite similar to Fig. 2a, in which one of the co-ordinated grains is larger and at least one of the connecting boundaries becomes curved. It is quite important to point out that the curved boundaries make the pores located on the three-grain junctions unstable, otherwise these pores would sit steadily with straight boundaries, i.e. in an equilibrium state, as shown in Fig. 4a. (Note: in some cases some straight boundaries of pores located on the three-grain junctions would disappear suddenly and make these pores two-sided, but it is believed that the disappearance of boundaries is due to the instability rather than to the growth of bubbles. In the second figure the simulated pore has become two-sided and lenticular, as shown in Fig. 2b. Because bubbles become unstable as time elapses, they sometimes disappear quite suddenly. Fortunately, one photograph of separtion of a simulated pore from the boundary was taken, and is shown in Fig. 3c, in which the simulated pore has become spherical. In Fig. 4, note that if the simulated pore indicated by the white arrow is maintained in an equilibrium state as shown in Fig. 4a, it would disappear rather than become a two-sided bubble (Fig. 4b).