We present SmartPhOx, a pure camera-andflashlight smartphone-based pulse oximetry solution. We build on the ratio-of-ratios (RR) method and linear regression, an elegant approach resting on the Beer-Lambert law and landing itself to efficient smartphone implementations. However, its implementations without specialized hardware have so far proved to be unsuitable for clinical use, in particular due to the instability of the RR measurements. We use an empirical study to shed light on the reasons why and propose using the very RR measurements to filter RR measurements -a new paradigm we call the Meta-Region of interest (Meta-ROI). We design a complete-system architecture, including a novel data structure for storing and RR values in the time and space dimensions and an efficient algorithm for identifying Meta-ROI. Results from an Android implementation of SmartPhOx with the participation of 37 volunteers show that it is the first pure camera-and-flashlight solution to meet the FDA requirement for Root Mean Square Error (RMSE).