Lithium ion (Li-ion) battery pack is a complex system consisting of numerous cells connected in parallel and series. The performance of the pack is highly dependent on the health of each individual in-pack cell. An overcharged or discharged cell connected in a parallel string could change the total capacity of the battery pack. In a pack, current-split estimation plays an important role to monitor the cell functions. Therefore, a scheme is required to estimate current-split accurately, which can thereby help to improve the overall pack performance. To what follows, a recursive weighted covariance-based estimation method (RWEM) was proposed to estimate the current-split of each set of parallel connected cells. RWEM assigns weights to the interconnected cell structure by using correlation information between battery parameters in order to estimate the current-split. This was achieved by first deriving the one-step prediction error method, where consistency for covariance was proved. Furthermore, iterative recursion for sparse measurements was also considered. Performance evaluations were conducted by analyzing sets of real-time measurements collected from Li-ion battery pack used in electric vehicles (EVs). Results show that the proposed filter accurately estimated the battery parameters even in the presence of faults and random-noise variances.