Turbine blade tip clearance varies in a three-dimensional (3-D) direction in an aero-engine. The traditional fiber-optic sensor can only measure the radial clearance, and the results are affected by the interdimensional coupling effect. In this paper, a sensor array consisting of three of the two-circle coaxial fiber bundles, arranged along the isosceles triangle layout, was used to realize the decoupling of blade tip clearance 3-D parameters. These parameters included radial clearance, axial deviation angle, and circumferential deviation angle from the sensor outputs, leveraging a BP neural network method. A special fiber-optic sensor array and signal conditioning circuit were developed, and the experimental tests for static and dynamic properties of the measurement system were performed. Experimental results indicate that the maximum error of radial clearance static measurement is 47 μm with a standard deviation of 10 μm. The maximum error of axial and circumferential deviation angle static measurements are 0.49° and 2.32° with standard deviations of 0.13° and 0.36°, respectively; the measurement system performed well in terms of repeatability and reliability. The dynamic measurement standard deviation of radial clearance is less than 18 μm with axial and circumferential deviation angles of less than 0.2° and 0.5°, respectively. This satisfies the demand of real-time detection of blade tip clearance 3-D parameters of turbine blades of an aero-engine.