Blockchain, a distributed ledger technology, can potentially be deployed in a wide range of applications. Among these applications, decentralized payment systems (e.g. Bitcoin) have been one of the most mature blockchain applications with widespread adoption. While the early designs (e.g. Bitcoin) are often the currency of choice by cybercriminals (e.g., in ransomware incidents), they only provide pseudo-anonymity, in the sense that anyone can deanonymize Bitcoin transactions by using information in the blockchain. To strengthen the privacy protection of decentralized payment systems, a number of solutions such as Monero and Zerocash have been proposed. However, completely Decentralized Anonymous Payment (DAP) systems can be criminally exploited, for example in online extortion and money laundering activities. Recognizing the importance of regulation, we present a novel definition of Decentralized Conditional Anonymous Payment (DCAP) and describe the corresponding security requirements. In order to construct a concrete DCAP system, we first design a Condition Anonymous Payment (CAP) scheme (based on our proposed signature of knowledge), whose security can be demonstrated under the defined formal semantic and security models. To demonstrate utility, we compare the performance of our proposal with that of Zerocash under the same parameters and testing environment.