Abnormal deposition of brain amyloid is a major hallmark of Alzheimer’s disease (AD). The toxic extracellular amyloid plaques originating from the aberrant aggregation of beta-amyloid (Aβ) protein are considered to be the major cause of clinical deficits such as memory loss and cognitive impairment. Two-photon excited fluorescence (TPEF) microscopy provides high spatial resolution, minimal invasiveness, and long-term monitoring capability. TPEF imaging of amyloid plaques in AD transgenic mice models has greatly facilitated studies of the AD pathological mechanism. However, the imaging of deep cortical layers is still hampered by the conventional amyloid probes with short excitation/emission wavelength. In this work, we report that a near-infrared (NIR) probe, named CRANAD-3, is far superior for deep in vivo TPEF imaging of brain amyloid in comparison with the commonly used short-wavelength probe. Our findings show that the major interference for TPEF signal of the NIR probe is from the autofluorescence of lipofuscin, the “aging-pigment” in the brain. To eliminate the interference, we characterized the lipofuscin fluorescence in the aged brains of AD mice and found that it has unique broad emission and short lifetime. The lipofuscin signal can be clearly separated from the fluorescence of CRANAD-3 and fluorescent protein via a ratio-based unmixing method. Our results demonstrate the great advantages of NIR probes for in vivo deep-tissue imaging of amyloid plaques in AD.