The ability to detect low numbers of microbial cells in food and clinical samples is highly valuable but remains a challenge. Here we present a detection system (called 'APC-Cas') that can detect very low numbers of a bacterial pathogen without isolation, using a three-stage amplification to generate powerful fluorescence signals. APC-Cas involves a combination of nucleic acid-based allosteric probes and CRISPR-Cas13a components. It can selectively and sensitively quantify Salmonella Enteritidis cells (from 1 to 105 CFU) in various types of samples such as milk, showing similar or higher sensitivity and accuracy compared with conventional real-time PCR. Furthermore, APC-Cas can identify low numbers of S. Enteritidis cells in mouse serum, distinguishing mice with early- and late-stage infection from uninfected mice. Our method may have potential clinical applications for early diagnosis of pathogens.