Femtosecond near-infrared (NIR, 1030 nm) and near-ultraviolet (NUV, 257 nm) laser ablation (fs-LA) were evaluated for U-Pb dating of zircons using inductively coupled plasma mass spectrometry (ICP-MS). The apparent ages are in agreement with the reference ages, within expanded combined uncertainties, at both wavelengths, but NIR fs-LA produces poorer uncertainties and a ≈ 3.5% bias on the reference age of Mudtank and Plešovice zircon. The time resolved 238U signal intensity rises sharply and progressively decreases in NUV, whereas patterns in NIR follow progressive increase and then decrease during ablation. Synchrotron X-ray micro-tomography (XRMT) imaging of an ablated quartz permitted to link the morphology and dimensions of fs-LA craters to the performances of fs-LA-ICP-MS analyses on zircons. Average ablation rates are similar at 3 J.cm-2 for both wavelengths. However, the ablation rate is higher in NIR at 7 J.cm-2 than in NUV at 2 J.cm-2, whereas the fs-LA-ICP-MS yield (in counts.shot-1.ppm-1) is lower in NIR at 7 J.cm-2 than in NUV at 2 J.cm-2. This paradox as well as the biased or imprecise fs-LA-ICP-MS analyses are explained by nonlinear shot-to-shot ablation in NIR compared to NUV, induced by: (i) catastrophic ablation, (ii) incubation effects, and, to a minor extent (iii) preferential ionization of U relative to Pb. Consequently, NIR fs-LA is capable of producing accurate U-Pb ages thanks to the properties of femtosecond laser matter interactions, but with a higher fluence, sample consumption and poorer statistics than those obtained with NUV fs-LA.