KRAS mutation is a common point mutation which occurs in ~30% of all human cancers. Early assessment of KRAS mutation status is critical for prediction of clinical treatment outcomes. However, current diagnostic methods are based on either polymerase-chain-reaction (PCR) or next-generation-sequencing (NGS) analysis of biopsy samples, which are complex, time consuming, and lack portability. Here, we report a cost-effective smartphone-based fluorescence microscopy platform for detection of KRAS point mutations by imaging targeted DNA sequencing reactions in preserved tumor slides. Smartphone-based KRAS mutation detection was conducted in two steps: 1) in situ rolling-circle-amplification (RCA) combined with ligation chemistry to label wild type/mutant strains with different fluorescent colors, and 2) rapidly scanning the sample by a smartphone microscope to quantify mutant-to-wild type ratios. This smartphone microscope contains two laser diodes (532 and 638 nm) for dual-color fluorescence detection (Cy3 & Cy5) and an additional white LED for brightfield imaging. We first imaged and analyzed synthetic or extracted DNA from model cell lines captured and amplified on glass slides. The smartphone fluorescence microscope was able to detect as low as 1fM target DNA sequence, and demonstrated a high sequencing depth (1:1000 mutant:wild type ratio), comparable to the sensitivity of FDA-approved KRAS PCR-based tests. Furthermore, the device was applied for in situ mutation detection in cell lines and real patient tumor slices. A machine learning algorithm was also developed to improve the recognition of target signals against the nonspecific background. Overall, smartphone-based in situ mutation detection resulted in 100% concordance to clinical NGS analysis.