diff --git a/specs/opcode/35CALLDATALOAD.md b/specs/opcode/35CALLDATALOAD.md
new file mode 100644
index 000000000..a27bc9d36
--- /dev/null
+++ b/specs/opcode/35CALLDATALOAD.md
@@ -0,0 +1,41 @@
+# CALLDATALOAD opcode
+
+## Procedure
+
+The `CALLDATALOAD` opcode gets input data of current environment.
+
+## EVM Behaviour
+
+Stack input is the byte offset to read call data from. Stack output is a 32-byte value starting from the given offset of the call data. All bytes after the end of the call data are set to `0`.
+
+## Constraints
+
+1. opId == 0x35
+2. State Transition:
+   - if is_root_call:
+     - rw_counter += 3 (1 stack read, 1 call context read, 1 stack write)
+   - if is_internal_call:
+     - rw_counter += rw_counter_offset ∈ {5, 6, ..., 36, 37} (1 stack read, 3 call context reads, i ∈ {0, 1, ..., 31, 32} memory reads, 1 stack write)
+   - stack_pointer unchanged
+   - pc + 1
+   - gas - 3
+3. Lookups:
+   - `offset` is at the top of the stack
+   - `tx_id` is in the RW table (call context)
+   - if is_root_call (where `src_addr = offset`):
+     - `calldata_length` is in the TX table
+     - i ∈ {0, 1, ..., 31, 32} lookups for `i in range(32)`: if `buffer.read_flag(i)` then the i'th byte of the element on top of the stack `calldata_word[i]` is in the TX table {tx id, call data, src_addr + i}
+   - if is_internal_call (where `src_addr = offset + calldata_offset`):
+     - `calldata_length` is in the RW table (call context)
+     - `calldata_offset` is in the RW table (call context)
+     - `caller_id` is in the RW table (call context)
+     - i ∈ {0, 1, ..., 31, 32} lookups for `i in range(32)`: if `buffer.read_flag(i)` then the i'th byte of the element on top of the stack `calldata_word[i]` is in the RW table {memory, src_addr + i, caller_id}
+
+## Exceptions
+
+1. Stack underflow: stack is empty, stack pointer = 1024
+2. Out of gas: remaining gas is not enough for this opcode
+
+## Code
+
+Please refer to `src/zkevm_specs/evm/execution/calldataload.py`.
diff --git a/src/zkevm_specs/evm/execution/__init__.py b/src/zkevm_specs/evm/execution/__init__.py
index be699cbec..6adff7800 100644
--- a/src/zkevm_specs/evm/execution/__init__.py
+++ b/src/zkevm_specs/evm/execution/__init__.py
@@ -15,6 +15,7 @@
 from .caller import *
 from .callvalue import *
 from .calldatacopy import *
+from .calldataload import *
 from .gas import *
 from .jump import *
 from .jumpi import *
@@ -35,6 +36,7 @@
     ExecutionState.CALLER: caller,
     ExecutionState.CALLVALUE: callvalue,
     ExecutionState.CALLDATACOPY: calldatacopy,
+    ExecutionState.CALLDATALOAD: calldataload,
     ExecutionState.CALLDATASIZE: calldatasize,
     ExecutionState.COINBASE: coinbase,
     ExecutionState.TIMESTAMP: timestamp,
diff --git a/src/zkevm_specs/evm/execution/calldataload.py b/src/zkevm_specs/evm/execution/calldataload.py
new file mode 100644
index 000000000..c2361cbc3
--- /dev/null
+++ b/src/zkevm_specs/evm/execution/calldataload.py
@@ -0,0 +1,57 @@
+from ..instruction import Instruction, Transition
+from ..opcode import Opcode
+from ..table import RW, CallContextFieldTag, TxContextFieldTag
+from ..util import BufferReaderGadget
+from ...util.param import N_BYTES_WORD
+
+
+def calldataload(instruction: Instruction):
+    opcode = instruction.opcode_lookup(True)
+    instruction.constrain_equal(opcode, Opcode.CALLDATALOAD)
+
+    # offset is the 64-bit offset to start reading 32-bytes from start of calldata.
+    offset = instruction.rlc_to_fq_exact(instruction.stack_pop(), n_bytes=8)
+
+    tx_id = instruction.call_context_lookup(CallContextFieldTag.TxId, RW.Read)
+
+    if instruction.curr.is_root:
+        calldata_length = instruction.tx_context_lookup(tx_id, TxContextFieldTag.CallDataLength)
+        calldata_offset = 0
+    else:
+        calldata_length = instruction.call_context_lookup(CallContextFieldTag.CallDataLength)
+        calldata_offset = instruction.call_context_lookup(CallContextFieldTag.CallDataOffset)
+        caller_id = instruction.call_context_lookup(CallContextFieldTag.CallerId)
+
+    src_addr = offset + calldata_offset
+    src_addr_end = calldata_length + calldata_offset
+
+    buffer_reader = BufferReaderGadget(
+        instruction, N_BYTES_WORD, src_addr, src_addr_end, N_BYTES_WORD
+    )
+
+    calldata_word = []
+    for idx in range(N_BYTES_WORD):
+        if buffer_reader.read_flag(idx):
+            if instruction.curr.is_root:
+                tx_byte = instruction.tx_calldata_lookup(tx_id, src_addr + idx)
+                buffer_reader.constrain_byte(idx, tx_byte)
+                calldata_word.append(int(tx_byte))
+            else:
+                mem_byte = instruction.memory_lookup(RW.Read, src_addr + idx, caller_id)
+                buffer_reader.constrain_byte(idx, mem_byte)
+                calldata_word.append(int(mem_byte))
+        else:
+            buffer_reader.constrain_byte(idx, 0)
+            calldata_word.append(0)
+
+    instruction.constrain_equal(
+        instruction.stack_push(),
+        instruction.bytes_to_rlc(bytes(calldata_word)),
+    )
+
+    instruction.step_state_transition_in_same_context(
+        opcode,
+        rw_counter=Transition.delta(instruction.rw_counter_offset),
+        program_counter=Transition.delta(1),
+        stack_pointer=Transition.same(),
+    )
diff --git a/src/zkevm_specs/evm/instruction.py b/src/zkevm_specs/evm/instruction.py
index d37b1c4a8..bf5e36a45 100644
--- a/src/zkevm_specs/evm/instruction.py
+++ b/src/zkevm_specs/evm/instruction.py
@@ -302,6 +302,9 @@ def word_to_lo_hi(self, word: RLC) -> Tuple[FQ, FQ]:
     def int_to_rlc(self, value: int, n_bytes: int) -> RLC:
         return RLC(value, self.randomness, n_bytes)
 
+    def bytes_to_rlc(self, value: bytes) -> RLC:
+        return RLC(value, self.randomness, len(value))
+
     def bytes_to_int(self, value: bytes) -> int:
         assert len(value) <= MAX_N_BYTES, "Too many bytes to composite an integer in field"
         return int.from_bytes(value, "little")
diff --git a/tests/evm/test_calldataload.py b/tests/evm/test_calldataload.py
new file mode 100644
index 000000000..a207217f4
--- /dev/null
+++ b/tests/evm/test_calldataload.py
@@ -0,0 +1,229 @@
+import pytest
+
+from typing import Optional
+from zkevm_specs.evm import (
+    Bytecode,
+    CallContextFieldTag,
+    ExecutionState,
+    RW,
+    RWTableTag,
+    StepState,
+    Tables,
+    Transaction,
+    verify_steps,
+)
+from zkevm_specs.util import rand_fp, RLC, U64
+
+TESTING_DATA = (
+    (
+        bytes.fromhex("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"),
+        0x20,
+        0x00,
+        bytes.fromhex("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"),
+        True,
+        None,
+    ),
+    (
+        bytes.fromhex("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"),
+        0x20,
+        0x1F,
+        bytes.fromhex("FF00000000000000000000000000000000000000000000000000000000000000"),
+        True,
+        None,
+    ),
+    (
+        bytes.fromhex("a1bacf5488bfafc33bad736db41f06866eaeb35e1c1dd81dfc268357ec98563f"),
+        0x20,
+        0x10,
+        bytes.fromhex("6eaeb35e1c1dd81dfc268357ec98563f00000000000000000000000000000000"),
+        True,
+        None,
+    ),
+    (
+        bytes.fromhex("a1bacf5488bfafc33bad736db41f06866eaeb35e1c1dd81dfc268357ec98563f"),
+        0x20,
+        0x10,
+        bytes.fromhex("6eaeb35e1c1dd81dfc268357ec98563f00000000000000000000000000000000"),
+        False,
+        0x00,
+    ),
+    (
+        bytes.fromhex("a1bacf5488bfafc33bad736db41f06866eaeb35e1c1dd81dfc268357ec98563fab"),
+        0x20,
+        0x10,
+        bytes.fromhex("aeb35e1c1dd81dfc268357ec98563fab00000000000000000000000000000000"),
+        False,
+        0x01,
+    ),
+)
+
+
+@pytest.mark.parametrize(
+    "call_data, call_data_length, offset, expected_stack_top, is_root, call_data_offset",
+    TESTING_DATA,
+)
+def test_calldataload(
+    call_data: bytes,
+    call_data_length: U64,
+    offset: U64,
+    expected_stack_top: bytes,
+    is_root: bool,
+    call_data_offset: Optional[U64],
+):
+    randomness = rand_fp()
+
+    tx = Transaction(id=1)
+    if is_root:
+        tx.call_data = call_data
+
+    offset_rlc = RLC(offset, randomness)
+    expected_stack_top = RLC(expected_stack_top, randomness)
+
+    bytecode = Bytecode().push(offset_rlc, n_bytes=32).calldataload().stop()
+    bytecode_hash = RLC(bytecode.hash(), randomness)
+
+    if is_root:
+        call_id = 1
+    else:
+        call_id = 2
+        parent_call_id = 1
+
+    rws = set(
+        [
+            (1, RW.Write, RWTableTag.Stack, call_id, 1023, 0, offset_rlc, 0, 0, 0),
+            (2, RW.Read, RWTableTag.Stack, call_id, 1023, 0, offset_rlc, 0, 0, 0),
+            (3, RW.Read, RWTableTag.CallContext, call_id, CallContextFieldTag.TxId, 0, 1, 0, 0, 0),
+        ]
+    )
+    if is_root:
+        rws.add((4, RW.Write, RWTableTag.Stack, call_id, 1023, 0, expected_stack_top, 0, 0, 0))
+        rw_counter_stop = 5
+    else:
+        # add to RW table call context, call data length (read)
+        rws.add(
+            (
+                4,
+                RW.Read,
+                RWTableTag.CallContext,
+                call_id,
+                CallContextFieldTag.CallDataLength,
+                0,
+                call_data_length,
+                0,
+                0,
+                0,
+            )
+        )
+        # add to RW table call context, call data offset (read)
+        rws.add(
+            (
+                5,
+                RW.Read,
+                RWTableTag.CallContext,
+                call_id,
+                CallContextFieldTag.CallDataOffset,
+                0,
+                call_data_offset,
+                0,
+                0,
+                0,
+            )
+        )
+        # add to RW table call context, caller'd ID (read)
+        rws.add(
+            (
+                6,
+                RW.Read,
+                RWTableTag.CallContext,
+                call_id,
+                CallContextFieldTag.CallerId,
+                0,
+                parent_call_id,
+                0,
+                0,
+                0,
+            )
+        )
+        rw_counter = 7
+        # add to RW table memory (read)
+        for i in range(0, len(call_data)):
+            idx = offset + call_data_offset + i
+            if idx < len(call_data):
+                rws.add(
+                    (
+                        rw_counter,
+                        RW.Read,
+                        RWTableTag.Memory,
+                        parent_call_id,
+                        idx,
+                        0,
+                        call_data[idx],
+                        0,
+                        0,
+                        0,
+                    )
+                )
+                rw_counter += 1
+        # add to RW table stack (write)
+        rws.add(
+            (
+                rw_counter,
+                RW.Write,
+                RWTableTag.Stack,
+                call_id,
+                1023,
+                0,
+                expected_stack_top,
+                0,
+                0,
+                0,
+            )
+        )
+        rw_counter_stop = rw_counter + 1
+
+    tables = Tables(
+        block_table=set(),
+        tx_table=set(tx.table_assignments(randomness)),
+        bytecode_table=set(bytecode.table_assignments(randomness)),
+        rw_table=rws,
+    )
+
+    verify_steps(
+        randomness=randomness,
+        tables=tables,
+        steps=[
+            StepState(
+                execution_state=ExecutionState.PUSH,
+                rw_counter=1,
+                call_id=call_id,
+                is_root=is_root,
+                is_create=False,
+                code_source=bytecode_hash,
+                program_counter=0,
+                stack_pointer=1024,
+                gas_left=6,
+            ),
+            StepState(
+                execution_state=ExecutionState.CALLDATALOAD,
+                rw_counter=2,
+                call_id=call_id,
+                is_root=is_root,
+                is_create=False,
+                code_source=bytecode_hash,
+                program_counter=33,
+                stack_pointer=1023,
+                gas_left=3,
+            ),
+            StepState(
+                execution_state=ExecutionState.STOP,
+                rw_counter=rw_counter_stop,
+                call_id=call_id,
+                is_root=is_root,
+                is_create=False,
+                code_source=bytecode_hash,
+                program_counter=34,
+                stack_pointer=1023,
+                gas_left=0,
+            ),
+        ],
+    )
diff --git a/tests/evm/test_calldatasize.py b/tests/evm/test_calldatasize.py
index 156f7bce0..3f8a1879a 100644
--- a/tests/evm/test_calldatasize.py
+++ b/tests/evm/test_calldatasize.py
@@ -3,7 +3,6 @@
 from zkevm_specs.evm import (
     ExecutionState,
     StepState,
-    Opcode,
     verify_steps,
     Tables,
     RWTableTag,