The Prolog primitive thread_cancel/1, which simply cancels a thread as recommended in ISO/IEC DTR 13211-5:2007, is conspicuously absent in well-maintained, widely used multithreaded Prolog systems. The ability to cancel a thread is useful for application development and is critical to Prolog embeddability. The difficulty of cancelling a thread is due to the instant mapping of Prolog multithreading primitives to the native-machine thread methods. This paper reports on an attempt to implement thread cancellation using self-blocking threads. A thread blocks at the same safe execution point where the state of the underlying virtual machine is defined. A blocked thread awaits a notification to resume or terminate. A resumed thread may be redirected to self-block by a blocking primitive. Experimental results based on a prototype implementation show that using self-blocking threads not only simplifies the implementation of thread cancellation but also improves the performance of message-passing primitives.