In daily life, a person’s gait—an important marker for his/her health status—is usually assessed using inertial sensors fixed to lower limbs or trunk. Such sensor locations are not well suited for continuous and long duration measurements. A better location would be the wrist but with the drawback of the presence of perturbative movements independent of walking. The aim of this study was to devise and validate an algorithm able to accurately estimate walking cadence and speed for daily life walking in various environments based on acceleration measured at the wrist. To this end, a cadence likelihood measure was designed, automatically filtering out perturbative movements and amplifying the periodic wrist movement characteristic of walking. Speed was estimated using a piecewise linear model. The algorithm was validated for outdoor walking in various and challenging environments (e.g., trail, uphill, downhill). Cadence and speed were successfully estimated for all conditions. Overall median (interquartile range) relative errors were −0.13% (−1.72 2.04%) for instantaneous cadence and −0.67% (−6.52 6.23%) for instantaneous speed. The performance was comparable to existing algorithms for trunk- or lower limb-fixed sensors. The algorithm’s low complexity would also allow a real-time implementation in a watch.