Resistive sensors based on metal oxide nanostructures are very interesting because they are tiny, simple, cheap to fabricate and to use, possibly low power consuming, and can sense almost any gas. Unfortunately, this last property is not so positive, because it means that these solid state sensors completely lack selectivity. This comes from the fact that their response is one-dimensional, a dimensionless ratio that indicates at most whether the gas is oxidizing or reducing. This negative aspect can be circumvented by combining the response in different conditions (for example at different working temperatures) so as to get more significant information. Indeed, combining N sensors working at different temperatures, we obtain an N-dimensional point that contains the information of N raw responses, but also the N*(N-1)/2 correlations between the responses themselves. Besides usual PCA visualization, this information can be used in different ways to achieve real smart detection: RGB coding gives a qualitative and semi-quantitative response (similar to a litmus paper), while machine learning algorithms make the system able to recognize the gas and estimate its concentration in a quantitative way. Both of these approaches make it possible to achieve a perfect classification of the gases tested and a very low error on the concentration estimate (5-15%), bridging the gap between simple resistive sensors and electronic noses. Tiny electronic noses that can be integrated into portable devices such as a smart phone are the first step towards a widespread network that will allow to monitor the quality of the air breathed by people and at the same time their health.