DNA sequencing in a lab-on-a-chip aims at providing cheap, high-speed analysis of low reagent volumes to, e.g., identify genomic deletions or insertions associated with genetic illnesses. Detecting single base-pair insertions/deletions from DNA fragments in the diagnostically relevant range of 150−1000 base-pairs requires a sizing accuracy of S < 10-3. Here we demonstrate S = 4×10-4. A microfluidic chip was post-processed by femtosecond-laser writing of an optical waveguide. 12 blue-labeled and 23 red-labeled DNA fragments were separated in size by capillary electrophoresis, each set excited by either of two lasers power-modulated at different frequencies, their fluorescence detected by a photomultiplier, and blue/red signals distinguished by Fourier analysis. Different calibration strategies were tested: a) use either set of DNA molecules as reference to calibrate the set-up and identify the base-pair sizes of the other set in the same flow experiment, thereby eliminating variations in temperature, wall-coating and sieving-gel conditions, and actuation voltages; b) use the same molecular set as reference and sample with the same fluorescence label, flown in consecutive experiments; c) perform cross-experiments based on different molecular sets with different labels, flown in consecutive experiments. From the results we conclude: Applying quadratic instead of linear fit functions improves the calibration accuracy. Blue-labeled molecules are separated with higher accuracy. The influence of dye label is higher than fluctuations between two experiments. Choosing a single, suitable dye label combined with reference calibration and sample investigation in consecutive experiments results in S = 4×10-4, enabling detection of single base-pair insertion/deletion in a lab-on-a-chip.