This paper presents a meaning-based method to spam filtering by distinguishing text without content from text with little content from text with normal content, based on the amount of meaning that can be automatically processed in the way humans do. The basic method assumes that a semantic analyzer will be able to produce less output from semantically less grammatical input text than from semantically well-formed text. The method was pilot-tested on a corpus of blog spam. Future improvements, including a method to distinguish semantically unified from semantically disparate text are sketched. The tested method, but even more the projected improvements, will open up the way to taking the spam filtering arms race to a new level very costly to spam producers.