This paper proposes an improved step detection approach for smartphones. The improvements consist in a new fuzzy-logic (FL) algorithm for user's step detection using accelerometer data. The proposed method exploits ranges of features reported in reference data on basic gait parameters. We also formulate robust defuzzification-rules related to future outputs. Furthermore, we compared the performances with different state-of-art detectors using the statistical criteria Precision, Recall and F1-Score. In real indoor scenarios, the proposed algorithm has exhibited robust step detection, with an improvement of F1-score up to 89.8% compared to traditional step detectors.