The craze‐initiation criteria for craze‐prone glassy polymers are examined with a model test method. In this method a prismatic sample of polystyrene with an embedded steel ball is subjected to a simple tension until crazing takes place around the interface. Stress analysis showed that various published craze‐initiation criteria will predict the appearance of the first crazes at different locations on the interface in such samples. Thus by comparing the experimental and analytical results it is possible to determine from the actual craze‐initiation locations which of the existing theories is likely to explain the craze initiation. The study revealed that the principal strain criterion gave the best prediction while the dilation criterion yielded the least likely result. The polystyrene samples with an embedded rubber ball were also tested and the data compared with the analytical results. In this case all criteria predict correctly that crazing will first occur at the equator of the rubber ball, indicating that the simple tension test of rubber ball samples is not suitable for determining the dominant craze‐initiation mechanism. Comparison of the crazing behavior in the rubber ball sample and the steel ball sample revealed that the crazes which developed around the rubber ball were more stable than those developed around the steel ball. The above observation may explain why harder inclusions are not as effective as rubber inclusions in toughening the glassy polymers. The incipient craze angles were examined and found to be in agreement with the general observation that the crazes form in the direction normal to the major principal stress.